Download Theory of Regression

Theory of Regression 1 The Course • 16 (or so) lessons – Some flexibility • Depends how we feel • What we get through 2 Part I: Theory of Regression 1. Models in statistics 2. Models with more than one parameter: regression 3. Why regression? 4. Samples to populations 5. Introducing multiple regression 6. More on multiple regression 3 Part 2: Application of regression 7. 8. 9. 10. 11. 12. Categorical predictor variables Assumptions in regression analysis Issues in regression analysis Non-linear regression Moderators (interactions) in regression Mediation and path analysis Part 3: Advanced Types of Regression 13. 14. 15. 16. Logistic Regression Poisson Regression Introducing SEM Introducing longitudinal multilevel models 4 House Rules • Jeremy must remember – Not to talk too fast • If you don’t understand – Ask – Any time • If you think I’m wrong – Ask. (I’m not always right) 5 Learning New Techniques • Best kind of data to learn a new technique – Data that you know well, and understand • Your own data – In computer labs (esp later on) – Use your own data if you like • My data – I’ll provide you with – Simple examples, small sample sizes • Conceptually simple (even silly) 6 Computer Programs • SPSS – Mostly • Excel – For calculations • • • • • GPower Stata (if you like) R (because it’s flexible and free) Mplus (SEM, ML?) AMOS (if you like) 7 8 9 Lesson 1: Models in statistics Models, parsimony, error, mean, OLS estimators 10 What is a Model? 11 What is a model? • Representation – Of reality – Not reality • Model aeroplane represents a real aeroplane – If model aeroplane = real aeroplane, it isn’t a model 12 • Statistics is about modelling – Representing and simplifying • Sifting – What is important from what is not important • Parsimony – In statistical models we seek parsimony – Parsimony  simplicity 13 Parsimony in Science • A model should be: – 1: able to explain a lot – 2: use as few concepts as possible • More it explains – The more you get • Fewer concepts – The lower the price • Is it worth paying a higher price for a better model? 14 A Simple Model • Height of five individuals – 1.40m – 1.55m – 1.80m – 1.62m – 1.63m • These are our DATA 15 A Little Notation Y Yi The (vector of) data that we are modelling The ith observation in our data. Y  4,5,6,7,8 Y2  5 16 Greek letters represent the true value in the population.  0 j  (Beta) Parameters in our model (population value) The value of the first parameter of our model in the population. The value of the jth parameter of our model, in the population. (Epsilon) The error in the population model. 17 Normal letters represent the values in our sample. These are sample statistics, which are used to estimate population parameters. b e Y A parameters in our model (sample statistics) The error in our sample. The data in our sample which we are trying to model. 18 Symbols on top change the meaning. Y The data in our sample which we are trying to model (repeated). ˆ Yi The estimated value of Y, for the ith case. Y The mean of Y. 19 So b1  ˆ1 I will use b1 (because it is easier to type)  20 • Not always that simple – some texts and computer programs use b = the parameter estimate (as we have used)  (beta) = the standardised parameter estimate SPSS does this. 21 A capital letter is the set (vector) of parameters/statistics B Set of all parameters (b0, b1, b2, b3 … bp) Rules are not used very consistently (even by me). Don’t assume you know what someone means, without checking. 22 • We want a model – To represent those data • Model 1: – 1.40m, 1.55m, 1.80m, 1.62m, 1.63m – Not a model • A copy – VERY unparsimonious • Data: 5 statistics • Model: 5 statistics – No improvement 23 • Model 2: – The mean (arithmetic mean) – A one parameter model n ˆ Yi  b0  Y   Yi i 1 n 24 • Which, because we are lazy, can be written as Y Y  n 25 The Mean as a Model 26 The (Arithmetic) Mean • We all know the mean – The ‘average’ – Learned about it at school – Forget (didn’t know) about how clever the mean is • The mean is: – An Ordinary Least Squares (OLS) estimator – Best Linear Unbiased Estimator (BLUE) 27 Mean as OLS Estimator • Going back a step or two • MODEL was a representation of DATA – We said we want a model that explains a lot – How much does a model explain? DATA = MODEL + ERROR ERROR = DATA - MODEL – We want a model with as little ERROR as possible 28 • What is error? Data (Y) Model (b0) mean Error (e) 1.40 -0.20 1.55 -0.05 1.80 1.60 0.20 1.62 0.02 1.63 0.03 29 • How can we calculate the ‘amount’ of error? • Sum of errors ERROR  ei  (Yi  Yˆ)  (Yi  b0 )   0.20   0.05  0.20  0.02  0.03 0 30 – 0 implies no ERROR • Not the case – Knowledge about ERROR is useful • As we shall see later 31 • Sum of absolute errors – Ignore signs ERROR   ei   Yi  Yˆ   Yi  b0  0.20  0.05  0.20  0.02  0.03  0.50 32 • Are small and large errors equivalent? – One error of 4 – Four errors of 1 – The same? – What happens with different data? • Y = (2, 2, 5) – b0 = 2 – Not very representative • Y = (2, 2, 4, 4) – b0 = any value from 2 - 4 – Indeterminate • There are an infinite number of solutions which would satisfy our criteria for minimum error 33 • Sum of squared errors (SSE) 2 i ERROR  e 2 ˆ  (Yi  Y )  (Yi  b0 ) 2   0.20   0.05  0.20  0.02  0.03 2 2 2 2 2  0.08 34 • Determinate – Always gives one answer • If we minimise SSE – Get the mean • Shown in graph – SSE plotted against b0 – Min value of SSE occurs when – b0 = mean 35 2 1.8 1.6 1.4 SSE 1.2 1 0.8 0.6 0.4 0.2 0 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 b0 36 The Mean as an OLS Estimate 37 Mean as OLS Estimate • The mean is an Ordinary Least Squares (OLS) estimate – As are lots of other things • This is exciting because – OLS estimators are BLUE – Best Linear Unbiased Estimators – Proven with Gauss-Markov Theorem • Which we won’t worry about 38 BLUE Estimators • Best – Minimum variance (of all possible unbiased estimators – Narrower distribution than other estimators • e.g. median, mode • Linear – Linear predictions Y Y – For the mean – Linear (straight, flat) line 39 • Unbiased – Centred around true (population) values – Expected value = population value – Minimum is biased. • Minimum in samples > minimum in population • Estimators – Errrmm… they are estimators • Also consistent – Sample approaches infinity, get closer to population values – Variance shrinks 40 SSE and the Standard Deviation • Tying up a loose end 2 ˆ SSE  (Yi  Y ) s 2 ˆ (Yi  Y ) n  2 ˆ (Yi  Y ) n 1 41 • SSE closely related to SD • Sample standard deviation – s – Biased estimator of population SD • Population standard deviation -  – Need to know the mean to calculate SD • Reduces N by 1 • Hence divide by N-1, not N – Like losing one df 42 Proof • That the mean minimises SSE – Not that difficult – As statistical proofs go • Available in – Maxwell and Delaney – Designing experiments and analysing data – Judd and McClelland – Data Analysis (out of print?) 43 What’s a df? • The number of parameters free to vary – When one is fixed • Term comes from engineering – Movement available to structures 44 0 df No variation available 1 df Fix 1 corner, the shape is fixed 45 Back to the Data • Mean has 5 (N) df – 1st moment •  has N –1 df – Mean has been fixed – 2nd moment – Can think of as amount cases vary away from the mean 46 While we are at it … • Skewness has N – 2 df – 3rd moment • Kurtosis has N – 3 df – 4rd moment – Amount cases vary from  47 Parsimony and df • Number of df remaining – Measure of parsimony • Model which contained all the data – Has 0 df – Not a parsimonious model • Normal distribution – Can be described in terms of mean and  • 2 parameters – (z with 0 parameters) 48 Summary of Lesson 1 • Statistics is about modelling DATA – Models have parameters – Fewer parameters, more parsimony, better • Models need to minimise ERROR – Best model, least ERROR – Depends on how we define ERROR – If we define error as sum of squared deviations from predicted value – Mean is best MODEL 49 50 51 Lesson 2: Models with one more parameter - regression 52 In Lesson 1 we said … • Use a model to predict and describe data – Mean is a simple, one parameter model 53 More Models Slopes and Intercepts 54 More Models • The mean is OK – As far as it goes – It just doesn’t go very far – Very simple prediction, uses very little information • We often have more information than that – We want to use more information than that 55 House Prices • In the UK, two of the largest lenders (Halifax and Nationwide) compile house price indices – Predict the price of a house – Examine effect of different circumstances • Look at change in prices – Guides legislation • E.g. interest rates, town planning 56 Predicting House Prices Beds £ (000s) 1 2 1 3 5 5 2 5 4 1 77 74 88 62 90 136 35 134 138 55 57 One Parameter Model • The mean Y  88.9 Yˆ  b0  Y SSE  11806.9 “How much is that house worth?” “£88,900” Use 1 df to say that 58 Adding More Parameters • We have more information than this – We might as well use it – Add a linear function of number of bedrooms (x1) Yˆ  b0  b1 x1 59 Alternative Expression • Estimate of Y (expected value of Y) Yˆ  b0  b1 x1 • Value of Y Yi  b0  b1 xi1  ei 60 Estimating the Model • We can estimate this model in four different, equivalent ways – Provides more than one way of thinking about it 1. 2. 3. 4. Estimating the slope which minimises SSE Examining the proportional reduction in SSE Calculating the covariance Looking at the efficiency of the predictions 61 Estimate the Slope to Minimise SSE 62 Estimate the Slope • Stage 1 – Draw a scatterplot – x-axis at mean • Not at zero • Mark errors on it – Called ‘residuals’ – Sum and square these to find SSE 63 160 140 120 100 80 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60 40 20 0 64 160 140 120 100 80 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60 40 20 0 65 • Add another slope to the chart – Redraw residuals – Recalculate SSE – Move the line around to find slope which minimises SSE • Find the slope 66 • First attempt: 67 • Any straight line can be defined with two parameters – The location (height) of the slope • b0 – Sometimes called a – The gradient of the slope • b1 68 • Gradient b1 units 1 unit 69 • Height b0 units 70 • Height • If we fix slope to zero – Height becomes mean – Hence mean is b0 • Height is defined as the point that the slope hits the y-axis – The constant – The y-intercept 71 • Why the constant? – b0x0 – Where x0 is 1.00 for every case • i.e. x0 is constant • Implicit in SPSS – Some packages force you to make it explicit – (Later on we’ll need to make it explicit) beds (x1) x0 1 1 2 1 1 1 3 1 5 1 5 1 2 1 5 1 4 1 1 1 £ (000s) 77 74 88 62 90 136 35 134 138 55 72 • Why the intercept? – Where the regression line intercepts the yaxis – Sometimes called y-intercept 73 Finding the Slope • How do we find the values of b0 and b1? – Start with we jiggle the values, to find the best estimates which minimise SSE – Iterative approach • Computer intensive – used to matter, doesn’t really any more • (With fast computers and sensible search algorithms – more on that later) 74 • Start with – b0=88.9 (mean) – b1=10 (nice round number) • SSE = 14948 – worse than it was – b0=86.9, – b0=66.9, – b0=56.9, – b0=46.9, – b0=51.9, – b0=51.9, – b0=46.9, – …….. b1=10, b1=10, b1=10, b1=10, b1=10, b1=12, b1=14, SSE=13828 SSE=7029 SSE=6628 SSE=8228 SSE=7178 SSE=6179 SSE=5957 75 • Quite a long time later – b0 = 46.000372 – b1 = 14.79182 – SSE = 5921 • Gives the position of the – Regression line (or) – Line of best fit • Better than guessing • Not necessarily the only method – But it is OLS, so it is the best (it is BLUE) 76 160 140 120 Price 100 80 60 Actual Price Predicted Price 40 20 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 Number of Bedrooms 77 • We now know – A house with no bedrooms is worth  £46,000 (??!) – Adding a bedroom adds  £15,000 • Told us two things – Don’t extrapolate to meaningless values of x-axis – Constant is not necessarily useful • It is necessary to estimate the equation 78 Standardised Regression Line • One big but: – Scale dependent • Values change – £ to €, inflation • Scales change – £, £000, £00? • Need to deal with this 79 • Don’t express in ‘raw’ units – Express in SD units – x1=1.72 – y=36.21 • b1 = 14.79 • We increase x1 by 1, and Ŷ increases by 14.79 14.79  (14.79 / 36.21)SDs  0.408SDs 80 • Similarly, 1 unit of x1 = 1/1.72 SDs – Increase x1 by 1 SD – Ŷ increases by 14.79  (1.72/1) = 8.60 • Put them both together b1   x1 y 81 14.79 1.72  0.706 36.21 • The standardised regression line – Change (in SDs) in Ŷ associated with a change of 1 SD in x1 • A different route to the same answer – Standardise both variables (divide by SD) – Find line of best fit 82 • The standardised regression line has a special name The Correlation Coefficient (r) (r stands for ‘regression’, but more on that later) • Correlation coefficient is a standardised regression slope – Relative change, in terms of SDs 83 Proportional Reduction in Error 84 Proportional Reduction in Error • We might be interested in the level of improvement of the model – How much less error (as proportion) do we have – Proportional Reduction in Error (PRE) • Mean only – Error(model 0) = 11806 • Mean + slope – Error(model 1) = 5921 85 ERROR(0)  ERROR(1) PRE  ERROR(0) ERROR(1) PRE  1  ERROR(0) 5921 PRE  1  11806 PRE  0.4984 86 • But we squared all the errors in the first place – So we could take the square root – (It’s a shoddy excuse, but it makes the point) 0.4984  0.706 • This is the correlation coefficient • Correlation coefficient is the square root of the proportion of variance explained 87 Standardised Covariance 88 Standardised Covariance • We are still iterating – Need a ‘closed-form’ – Equation to solve to get the parameter estimates • Answer is a standardised covariance – A variable has variance – Amount of ‘differentness’ • We have used SSE so far 89 • SSE varies with N – Higher N, higher SSE • Divide by N – Gives SSE per person – (Actually N – 1, we have lost a df to the mean) • The variance • Same as SD2 – We thought of SSE as a scattergram • Y plotted against X – (repeated image follows) 90 160 140 120 100 80 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60 40 20 0 91 • Or we could plot Y against Y – Axes meet at the mean (88.9) – Draw a square for each point – Calculate an area for each square – Sum the areas • Sum of areas – SSE • Sum of areas divided by N – Variance 92 Plot of Y against Y 180 160 140 120 100 0 20 40 60 80 80 100 120 140 160 180 60 40 20 0 93 Draw Squares 180 Area = 40.1 x 40.1 = 1608.1 160 138 – 88.9 = 40.1 140 138 – 88.9 = 40.1 120 100 0 20 35 – 88.9 = -53.9 40 60 80 80 100 120 140 160 180 60 40 35 – 88.9 = -53.9 20 Area = -53.9 x -53.9 = 2905.21 0 94 • What if we do the same procedure – Instead of Y against Y – Y against X • • • • Draw rectangles (not squares) Sum the area Divide by N - 1 This gives us the variance of x with y – The Covariance – Shortened to Cov(x, y) 95 96 Area = (-33.9) x (-2) = 67.8 55 – 88.9 = -33.9 4-3=1 138-88.9 = 49.1 1 - 3 = -2 Area = 49.1 x 1 = 49.1 97 • More formally (and easily) • We can state what we are doing as an equation – Where Cov(x, y) is the covariance ( x  x )( y  y ) Cov( x, y )  N 1 • Cov(x,y)=44.2 • What do points in different sectors do to the covariance? 98 • Problem with the covariance – Tells us about two things – The variance of X and Y – The covariance • Need to standardise it – Like the slope • Two ways to standardise the covariance – Standardise the variables first • Subtract from mean and divide by SD – Standardise the covariance afterwards 99 • First approach – Much more computationally expensive • Too much like hard work to do by hand – Need to standardise every value • Second approach – Much easier – Standardise the final value only • Need the combined variance – Multiply two variances – Find square root (were multiplied in first place) 100 • Standardised covariance  Cov( x , y ) Var( x )  Var( y )  44.2 2.9  1311  0.706 101 • The correlation coefficient – A standardised covariance is a correlation coefficient r Covariance  variance  variance  102 • Expanded … r  ( x  x )( y  y )    N 1   2 2  ( x  x ) ( y  y )     N 1   N 1 103 • This means … – We now have a closed form equation to calculate the correlation – Which is the standardised slope – Which we can use to calculate the unstandardised slope 104 We know that: r b1   x1 y We know that: b1  r  y x 1 105 b1  r  y x 1 0.706  36.21 b1  1.72 b1  14.79 • So value of b1 is the same as the iterative approach 106 • The intercept – Just while we are at it • The variables are centred at zero – We subtracted the mean from both variables – Intercept is zero, because the axes cross at the mean 107 • Add mean of y to the constant – Adjusts for centring y • Subtract mean of x – But not the whole mean of x – Need to correct it for the slope c  y  b1 x1 c  88.9  14.8  3 c  46.00 • Naturally, the same 108 Accuracy of Prediction 109 One More (Last One) • We have one more way to calculate the correlation – Looking at the accuracy of the prediction • Use the parameters – b0 and b1 – To calculate a predicted value for each case 110 Beds 1 2 1 3 5 5 2 5 4 1 Actual Predicted Price Price 77 60.80 74 75.59 88 60.80 62 90.38 90 119.96 136 119.96 35 75.59 134 119.96 138 105.17 55 60.80 • Plot actual price against predicted price – From the model 111 140 Predicted Value 120 100 80 60 40 20 20 40 60 80 100 Actual Value 120 140 160 112 • r = 0.706 – The correlation • Seems a futile thing to do – And at this stage, it is – But later on, we will see why 113 Some More Formulae • For hand calculation r xy x 2 y 2 • Point biserial  M r y1  M y 0  PQ sd y 114 • Phi (f) – Used for 2 dichotomous variables Vote P Vote Q Homeowner A: 19 B: 54 Not homeowner C: 60 D:53 BC  AD r ( A  B)(C  D)( A  C )( B  D) 115 • Problem with the phi correlation – Unless Px= Py (or Px = 1 – Py) • Maximum (absolute) value is < 1.00 • Tetrachoric can be used • Rank (Spearman) correlation – Used where data are ranked 6d r 2 n(n  1) 2 116 Summary • Mean is an OLS estimate – OLS estimates are BLUE • Regression line – Best prediction of DV from IV – OLS estimate (like mean) • Standardised regression line – A correlation 117 • Four ways to think about a correlation – 1. – 2. – 3. – 4. Standardised regression line Proportional Reduction in Error (PRE) Standardised covariance Accuracy of prediction 118 119 120 Lesson 3: Why Regression? A little aside, where we look at why regression has such a curious name. 121 Regression The or an act of regression; reversion; return towards the mean; return to an earlier stage of development, as in an adult’s or an adolescent’s behaving like a child (From Latin gradi, to go) • So why name a statistical technique which is about prediction and explanation? 122 • Francis Galton – Charles Darwin’s cousin – Studying heritability • Tall fathers have shorter sons • Short fathers have taller sons – ‘Filial regression toward mediocrity’ – Regression to the mean 123 • Galton thought this was biological fact – Evolutionary basis? • Then did the analysis backward – Tall sons have shorter fathers – Short sons have taller fathers • Regression to the mean – Not biological fact, statistical artefact 124 Other Examples • Secrist (1933): The Triumph of Mediocrity in Business • Second albums often tend to not be as good as first • Sequel to a film is not as good as the first one • ‘Curse of Athletics Weekly’ • Parents think that punishing bad behaviour works, but rewarding good behaviour doesn’t 125 Pair Link Diagram • An alternative to a scatterplot x y 126 r=1.00 x x x x x x x 127 r=0.00 x x x x x 128 From Regression to Correlation • Where do we predict an individual’s score on y will be, based on their score on x? – Depends on the correlation • r = 1.00 – we know exactly where they will be • r = 0.00 – we have no idea • r = 0.50 – we have some idea 129 r=1.00 Starts here Will end up here x y 130 r=0.00 Starts here Could end anywhere here x y 131 r=0.50 Probably end somewhere here Starts here x y 132 Galton Squeeze Diagram • Don’t show individuals – Show groups of individuals, from the same (or similar) starting point – Shows regression to the mean 133 r=0.00 Ends here Group starts here x Group starts here y 134 r=0.50 x y 135 r=1.00 x y 136 1 unit r units x y • Correlation is amount of regression that doesn’t occur 137 • No regression • r=1.00 x y 138 • Some regression • r=0.50 x y 139 r=0.00 • Lots (maximum) regression • r=0.00 x y 140 Formula zˆ y  rxy z x 141 Conclusion • Regression towards mean is statistical necessity regression = perfection – correlation • Very non-intuitive • Interest in regression and correlation – From examining the extent of regression towards mean – By Pearson – worked with Galton – Stuck with curious name • See also Paper B3 142 143 144 Lesson 4: Samples to Populations – Standard Errors and Statistical Significance 145 The Problem • In Social Sciences – We investigate samples • Theoretically – Randomly taken from a specified population – Every member has an equal chance of being sampled – Sampling one member does not alter the chances of sampling another • Not the case in (say) physics, biology, etc. 146 Population • But it’s the population that we are interested in – Not the sample – Population statistic represented with Greek letter – Hat means ‘estimate’ ˆ b x  ˆx 147 • Sample statistics (e.g. mean) estimate population parameters • Want to know – Likely size of the parameter – If it is > 0 148 Sampling Distribution • We need to know the sampling distribution of a parameter estimate – How much does it vary from sample to sample • If we make some assumptions – We can know the sampling distribution of many statistics – Start with the mean 149 Sampling Distribution of the Mean • Given – Normal distribution – Random sample – Continuous data • Mean has a known sampling distribution – Repeatedly sampling will give a known distribution of means – Centred around the true (population) mean () 150 Analysis Example: Memory • Difference in memory for different words – 10 participants given a list of 30 words to learn, and then tested – Two types of word • Abstract: e.g. love, justice • Concrete: e.g. carrot, table 151 Concrete Abstract 12 4 11 7 4 6 9 12 8 6 12 10 9 8 8 5 12 10 8 4 Diff (x) 8 4 -2 -3 2 2 1 3 2 4 x  2.1  x  3.11 N  10 152 Confidence Intervals • This means – If we know the mean in our sample – We can estimate where the mean in the population () is likely to be • Using – The standard error (se) of the mean – Represents the standard deviation of the sampling distribution of the mean 153 1 SD contains 68% Almost 2 SDs contain 95% 154 • We know the sampling distribution of the mean – t distributed – Normal with large N (>30) • Know the range within means from other samples will fall – Therefore the likely range of  x se( x )  n 155 • Two implications of equation – Increasing N decreases SE • But only a bit – Decreasing SD decreases SE • Calculate Confidence Intervals – From standard errors • 95% is a standard level of CI – 95% of samples the true mean will lie within the 95% CIs – In large samples: 95% CI = 1.96  SE – In smaller samples: depends on t distribution (df=N-1=9) 156 x  2.1,  x  3.11, N  10  x 3.11 se( x )    0.98 n 10 157 95% CI  2.26  0.98  2.22 x  CI    x  CI -0.12    4.32 158 What is a CI? • (For 95% CI): • 95% chance that the true (population) value lies within the confidence interval? • 95% of samples, true mean will land within the confidence interval? 159 Significance Test • Probability that  is a certain value – Almost always 0 • Doesn’t have to be though • We want to test the hypothesis that the difference is equal to 0 – i.e. find the probability of this difference occurring in our sample IF =0 – (Not the same as the probability that =0) 160 • Calculate SE, and then t – t has a known sampling distribution – Can test probability that a certain value is included x t se(x ) 2.1 t  2.14 0.98 p  0.061 161 Other Parameter Estimates • Same approach – Prediction, slope, intercept, predicted values – At this point, prediction and slope are the same • Won’t be later on • We will look at one predictor only – More complicated with > 1 162 Testing the Degree of Prediction • Prediction is correlation of Y with Ŷ – The correlation – when we have one IV • Use F, rather than t • Started with SSE for the mean only – This is SStotal – Divide this into SSresidual – SSregression • SStot = SSreg + SSres 163 F SSreg df1 SS res df 2 df1  k df 2  N  k  1, 164 • Back to the house prices – Original SSE (SStotal) = 11806 – SSresidual = 5921 • What is left after our model – SSregression = 11806 – 5921 = 5885 • What our model explains • Slope = 14.79 • Intercept = 46.0 • r = 0.706 165 F SSreg df1 SS res df 2 5885 1 F  7.95 5921 (10  1  1) df1  k  1 df 2  N  k  1  8 166 • F = 7.95, df = 1, 8, p = 0.02 – Can reject H0 • H0: Prediction is not better than chance – A significant effect 167 Statistical Significance: What does a p-value (really) mean? 168 A Quiz • Six questions, each true or false • Write down your answers (if you like) • An experiment has been done. Carried out perfectly. All assumptions perfectly satisfied. Absolutely no problems. • P = 0.01 – Which of the following can we say? 169 1. You have absolutely disproved the null hypothesis (that is, there is no difference between the population means). 170 2. You have found the probability of the null hypothesis being true. 171 3. You have absolutely proved your experimental hypothesis (that there is a difference between the population means). 172 4. You can deduce the probability of the experimental hypothesis being true. 173 5. You know, if you decide to reject the null hypothesis, the probability that you are making the wrong decision. 174 6. You have a reliable experimental finding in the sense that if, hypothetically, the experiment were repeated a great number of times, you would obtain a significant result on 99% of occasions. 175 OK, What is a p-value • Cohen (1994) “[a p-value] does not tell us what we want to know, and we so much want to know what we want to know that, out of desperation, we nevertheless believe it does” (p 997). 176 OK, What is a p-value • Sorry, didn’t answer the question • It’s The probability of obtaining a result as or more extreme than the result we have in the study, given that the null hypothesis is true • Not probability the null hypothesis is true 177 A Bit of Notation • Not because we like notation – But we have to say a lot less • • • • Probability – P Null hypothesis is true – H Result (data) – D Given - | 178 What’s a P Value • P(D|H) – Probability of the data occurring if the null hypothesis is true • Not • P(H|D) – Probability that the null hypothesis is true, given that we have the data = p(H) • P(H|D) ≠ P(D|H) 179 • What is probability you are prime minister – Given that you are british – P(M|B) – Very low • What is probability you are British – Given you are prime minister – P(B|M) – Very high • P(M|B) ≠ P(B|M) 180 • There’s been a murder – Someone bumped off a statto for talking too much • The police have DNA • The police have your DNA – They match(!) • DNA matches 1 in 1,000,000 people • What’s the probability you didn’t do the murder, given the DNA match (H|D) 181 • Police say: – P(D|H) = 1/1,000,000 • Luckily, you have Jeremy on your defence team • We say: – P(D|H) ≠ P(H|D) • Probability that someone matches the DNA, who didn’t do the murder – Incredibly high 182 Back to the Questions • Haller and Kraus (2002) – Asked those questions of groups in Germany – Psychology Students – Psychology lecturers and professors (who didn’t teach stats) – Psychology lecturers and professors (who did teach stats) 183 1. You have absolutely disproved the null hypothesis (that is, there is no difference between the population means). • True • • • • • 34% of students 15% of professors/lecturers, 10% of professors/lecturers teaching statistics False We have found evidence against the null hypothesis 184 2. You have found the probability of the null hypothesis being true. – 32% of students – 26% of professors/lecturers – 17% of professors/lecturers teaching statistics • • False We don’t know 185 3. You have absolutely proved your experimental hypothesis (that there is a difference between the population means). – – – • 20% of students 13% of professors/lecturers 10% of professors/lecturers teaching statistics False 186 4. You can deduce the probability of the experimental hypothesis being true. – 59% of students – 33% of professors/lecturers – 33% of professors/lecturers teaching statistics • False 187 5. You know, if you decide to reject the null hypothesis, the probability that you are making the wrong decision. • • • • • 68% of students 67% of professors/lecturers 73% of professors professors/lecturers teaching statistics False Can be worked out – P(replication) 188 6. You have a reliable experimental finding in the sense that if, hypothetically, the experiment were repeated a great number of times, you would obtain a significant result on 99% of occasions. – 41% of students – 49% of professors/lecturers – 37% of professors professors/lecturers teaching statistics • • False Another tricky one – It can be worked out 189 One Last Quiz • I carry out a study – All assumptions perfectly satisfied – Random sample from population – I find p = 0.05 • You replicate the study exactly – What is probability you find p < 0.05? 190 • I carry out a study – All assumptions perfectly satisfied – Random sample from population – I find p = 0.01 • You replicate the study exactly – What is probability you find p < 0.05? 191 • Significance testing creates boundaries and gaps where none exist. • Significance testing means that we find it hard to build upon knowledge – we don’t get an accumulation of knowledge 192 • Yates (1951) "the emphasis given to formal tests of significance ... has resulted in ... an undue concentration of effort by mathematical statisticians on investigations of tests of significance applicable to problems which are of little or no practical importance ... and ... it has caused scientific research workers to pay undue attention to the results of the tests of significance ... and too little to the estimates of the magnitude of the effects they are investigating 193 Testing the Slope • Same idea as with the mean – Estimate 95% CI of slope – Estimate significance of difference from a value (usually 0) • Need to know the sd of the slope – Similar to SD of the mean 194 s y. x  s y.x  s y. x 2 ˆ (Y  Y ) N  k 1 SSres N  k 1 5921   27.2 8 195 • Similar to equation for SD of mean • Then we need standard error - Similar (ish) • When we have standard error – Can go on to 95% CI – Significance of difference 196 se(by. x )  s y. x ( x  x ) 2 27.2 se(by. x )   5.24 26.9 197 • Confidence Limits • 95% CI – t dist with N - k - 1 df is 2.31 – CI = 5.24  2.31 = 12.06 • 95% confidence limits 14.8  12.1    14.8  12.1 2.7    26.9 198 • Significance of difference from zero – i.e. probability of getting result if =0 • Not probability that  = 0 b 14.7 t   2.81 se(b) 5.2 df  N  k  1  8 p  0.02 • This probability is (of course) the same as the value for the prediction 199 Testing the Standardised Slope (Correlation) • Correlation is bounded between –1 and +1 – Does not have symmetrical distribution, except around 0 • Need to transform it – Fisher z’ transformation – approximately normal z  0.5[ln( 1  r )  ln( 1  r )] 1 SE z  n3 200 z  0.5[ln( 1  0.706)  ln( 1  0.706)] z  0.879 1 1 SEz    0.38 n3 10  3 • 95% CIs – 0.879 – 1.96 * 0.38 = 0.13 – 0.879 + 1.96 * 0.38 = 1.62 201 • Transform back to correlation e 1 r  2y e 1 2y • 95% CIs = 0.13 to 0.92 • Very wide – Small sample size – Maybe that’s why CIs are not reported? 202 Using Excel • Functions in excel – Fisher() – to carry out Fisher transformation – Fisherinv() – to transform back to correlation 203 The Others • Same ideas for calculation of CIs and SEs for – Predicted score – Gives expected range of values given X • Same for intercept – But we have probably had enough 204 Lesson 5: Introducing Multiple Regression 205 Residuals • We said Y = b0 + b1x1 • We could have said Yi = b0 + b1xi1 + ei • We ignored the i on the Y • And we ignored the ei – It’s called error, after all • But it isn’t just error – Trying to tell us something 206 What Error Tells Us • Error tells us that a case has a different score for Y than we predict – There is something about that case • Called the residual – What is left over, after the model • Contains information – Something is making the residual  0 – But what? 207 160 140 swimming pool 120 Price 100 80 Unpleasant neighbours 60 Actual Price Predicted Price 40 20 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 Number of Bedrooms 208 • The residual (+ the mean) is the value of Y If all cases were equal on X • It is the value of Y, controlling for X • Other words: – Holding constant – Partialling – Residualising – Conditioned on 209 Beds £ (000s)Pred 1 2 1 3 5 5 2 5 4 1 77 74 88 62 90 136 35 134 138 55 61 76 61 90 120 120 76 120 105 61 Adj. Value Res 105 -16 90 2 62 -27 117 28 119 30 73 -16 129 41 75 -14 56 -33 95 6 210 • Sometimes adjustment is enough on its own – Measure performance against criteria • Teenage pregnancy rate – Measure pregnancy and abortion rate in areas – Control for socio-economic deprivation, and anything else important – See which areas have lower teenage pregnancy and abortion rate, given same level of deprivation • Value added education tables – Measure school performance – Control for initial intake 211 Control? • In experimental research – Use experimental control – e.g. same conditions, materials, time of day, accurate measures, random assignment to conditions • In non-experimental research – Can’t use experimental control – Use statistical control instead 212 Analysis of Residuals • What predicts differences in crime rate – After controlling for socio-economic deprivation – Number of police? – Crime prevention schemes? – Rural/Urban proportions? – Something else • This is what regression is about 213 • Exam performance – Consider number of books a student read (books) – Number of lectures (max 20) a student attended (attend) • Books and attend as IV, grade as DV 214 Book s Attend 0 1 0 2 4 4 1 4 3 0 9 15 10 16 10 20 11 20 15 15 Grade 45 57 45 51 65 88 44 87 89 59 First 10 cases 215 • Use books as IV – R=0.492, F=12.1, df=1, 28, p=0.001 – b0=52.1, b1=5.7 – (Intercept makes sense) • Use attend as IV – R=0.482, F=11.5, df=1, 38, p=0.002 – b0=37.0, b1=1.9 – (Intercept makes less sense) 216 100 90 80 70 Grade (100) 60 50 40 30 -1 0 1 2 3 4 5 Books 217 100 90 80 70 60 Grade 50 40 30 5 7 9 11 13 15 17 19 21 Attend 218 Problem • Use R2 to give proportion of shared variance – Books = 24% – Attend = 23% • So we have explained 24% + 23% = 47% of the variance – NO!!!!! 219 • Look at the correlation matrix BOOKS 1 ATTEND 0.44 1 GRADE 0.49 0.48 1 BOOKS ATTEND GRADE • Correlation of books and attend is (unsurprisingly) not zero – Some of the variance that books shares with grade, is also shared by attend 220 • I have access to 2 cars • My wife has access to 2 cars – We have access to four cars? – No. We need to know how many of my 2 cars are the same cars as her 2 cars • Similarly with regression – But we can do this with the residuals – Residuals are what is left after (say) books – See of residual variance is explained by attend – Can use this new residual variance to calculate SSres, SStotal and SSreg 221 • Well. Almost. – This would give us correct values for SS – Would not be correct for slopes, etc • Assumes that the variables have a causal priority – Why should attend have to take what is left from books? – Why should books have to take what is left by attend? • Use OLS again 222 • Simultaneously estimate 2 parameters – b1 and b2 – Y = b0 + b1x1 + b2x2 – x1 and x2 are IVs • Not trying to fit a line any more – Trying to fit a plane • Can solve iteratively – Closed form equations better – But they are unwieldy 223 3D scatterplot (2points only) y x2 x1 224 b2 y b1 b0 x2 x1 225 (Really) Ridiculous Equations 2   y  y x1  x1 x2  x2     y  y x2  x2 x1  x1 x2  x2  b1  2 2 2 x1  x1  x2  x2   x1  x1 x2  x2  2   y  y x2  x2 x1  x1     y  y x1  x1 x2  x2 x1  x1  b2  2 2 2 x2  x2  x1  x1   x2  x2 x1  x1  b0  y  b1 x1  b2 x2 226 • The good news – There is an easier way • The bad news – It involves matrix algebra • The good news – We don’t really need to know how to do it • The bad news – We need to know it exists 227 A Quick Guide to Matrix Algebra (I will never make you do it again) 228 Very Quick Guide to Matrix Algebra • Why? – Matrices make life much easier in multivariate statistics – Some things simply cannot be done without them – Some things are much easier with them • If you can manipulate matrices – you can specify calculations v. easily – e.g. AA’ = sum of squares of a column • Doesn’t matter how long the column 229 • A scalar is a number A scalar: 4 • A vector is a row or column of numbers A row vector: A column vector: 2 4 8 7 5   11  230 • A vector is described as rows x columns 2 4 8 7 – Is a 1  4 vector 5   11  – Is a 2  1 vector – A number (scalar) is a 1  1 vector 231 • A matrix is a rectangle, described as rows x columns 2 6 5 7 8   4 5 7 5 3 1 5 2 7 8   • Is a 3 x 5 matrix • Matrices are referred to with bold capitals - A is a matrix 232 • Correlation matrices and covariance matrices are special – They are square and symmetrical – Correlation matrix of books, attend and grade  1.00 0.44 0.49     0.44 1.00 0.48   0.49 0.48 1.00    233 • Another special matrix is the identity matrix I – A square matrix, with 1 in the diagonal and 0 in the off-diagonal 1  0 I 0  0 0 0 0  1 0 0 0 1 0  0 0 1 – Note that this is a correlation matrix, with correlations all = 0 234 Matrix Operations • Transposition – A matrix is transposed by putting it on its side A  7 5 6  – Transpose of A is A’ 7   A'   5  6   235 • Matrix multiplication – A matrix can be multiplied by a scalar, a vector or a matrix – Not commutative – AB  BA – To multiply AB • Number of rows in A must equal number of columns in B 236 • Matrix by vector a b  c  d e f g   j   aj  dk  gl       h    k    bj  ek  hl       i   l   cj  fk  il   2 3 5  2   33   2 3 5  2   4  9  20   43   7 7 1111 1313  331499 33  52    90   17 19 23   4 141              17 19 23   4   34  57  92   183  237 • Matrix by matrix a b  e     c d g f   ae  cf af  bh      h   ce  dg cf  dh   2 3   2 3   4  12 6  15           5 7   4 5  10  28 15  35   16 21     38 50  238 • Multiplying by the identity matrix – Has no effect – Like multiplying by 1 AI  A 2 3 1 0 2 3         5 7   0 1  5 7  239 • The inverse of J is: 1/J • J x 1/J = 1 • Same with matrices – Matrices have an inverse – Inverse of A is A-1 – AA-1=I • Inverting matrices is dull – We will do it once – But first, we must calculate the determinant 240 • The determinant of A is |A| • Determinants are important in statistics – (more so than the other matrix algebra) • We will do a 2x2 – Much more difficult for larger matrices 241 a b  A   c d A  ad  cb  1.0 0.3  A     0.3 1.0  A  1  1  0.3  0.3 A  0.91 242 • Determinants are important because – Needs to be above zero for regression to work – Zero or negative determinant of a correlation/covariance matrix means something wrong with the data • Linear redundancy • Described as: – Not positive definite – Singular (if determinant is zero) • In different error messages 243 • Next, the adjoint a b  A    c d   d  b adj A     c a  •Now 1 A  adj A A 1 244 • Find A-1  1.0 0.3  A     0.3 1.0  A  0.91 A A 1 1  1.0  0.3  1     0.91   0.3 1.0   1.10  0.33       0.33 1.10  245 Matrix Algebra with Correlation Matrices 246 Determinants • Determinant of a correlation matrix – The volume of ‘space’ taken up by the (hyper) sphere that contains all of the points  1.0 0.0  A   0.0 1.0  A  1.0 247 X X X X X  1.0 0.0  A   0.0 1.0  A  1.0 248 X X X  1.0 1.0  A   1.0 1.0  A  0.0 249 Negative Determinant • Points take up less than no space – Correlation matrix cannot exist – Non-positive definite matrix 250 Sometimes Obvious 1.0 1.2   A   1 . 2 1 . 0   A  0.44 251 Sometimes Obvious (If You Think) 0.9 0.9   1   A   0.9 1 0.9   0.9 0.9  1   A  2.88 252 Sometimes No Idea  1.00 0.76 0.40    A   0.76 1 0.30   0.40 0.30  1   A  0.01  1.00 0.75 0.40    A   0.75 1 0.30   0.40 0.30  1   A  0.0075 253 Multiple R for Each Variable • Diagonal of inverse of correlation matrix – Used to calculate multiple R – Call elements aij Ri .123...k 1  1 aii 254 Regression Weights • Where i is DV • j is IV bi . j  aij aij 255 Back to the Good News • We can calculate the standardised parameters as B=Rxx-1 x Rxy • Where – B is the vector of regression weights – Rxx-1 is the inverse of the correlation matrix of the independent (x) variables – Rxy is the vector of correlations of the correlations of the x and y variables – Now do exercise 3.2 256 One More Thing • The whole regression equation can be described with matrices – very simply Y  XB  E 257 • Where – Y = vector of DV – X = matrix of IVs – B = vector of coefficients • Go all the way back to our example 258 1 1  1  1 1  1 1  1 1  1 0 1 0 2 4 4 1 4 3 0 9  e1   45   e   57  5   2    e3   45  10       e4   51  16    b0  10     e5   65    b1        20     e6   88  b2    e7   44   11      20   e8   87   e   89  15   9     15   e10   59  259 1  1 1  1 1  1  1 1  1 1  0 1 0 2 4 4 1 4 3 0 The constant – literally a constant. Could be any  e1   45     but number,  it is most  e2  to  57make  convenient it 1. Used  e   45  3 to ‘capture’   the  intercept. 9  5 10   16   e4   51   b0       e5 10   65  b1        20    e6   88   b2   e    11   7   44   e8   87  20       15   e9   89   e   59  15   10    260 1  1 1  1 1  1  1 1  1 1  0 1 0 2 4 4 1 4 3 0 9  e1   45       5  e2   57   e   45  10   3    e4   51 16  The matrix ofvalues for  b0       and attend) e5 10  IVs (books 65  b1        20    e6   88   b2   e    11   7   44   e8   87  20       15   e9   89   e   59  15   10    261  e1   45  1 0 9         e2   57  1 1 5   e   45  1 0 10   3      e4   51  1 2 16  1 4 10  b0   e   65  The parameter   b1    5     4 20    e6   88  1 are estimates. We b2         trying to find the 1 1best 11   e7   44  values of these.  e8   87  1 4 20         e9   89  1 3 15  1 0 15   e   59     10    262 Error. We are trying to 0 9 1 this minimise   1 1  1 1  1  1 1  1 1  1 0 2 4 4 1 4 3 0  e1   45      5  e2   57   e   45  10   3    16   e4   51   b0       e5 10   65  b1        20    e6   88   b2   e    11   7   44   e8   87  20       15   e9   89   e   59  15   10    263 1 0  1 1 1 0  1 2 1 4  1 4  1 1 1 4  1 3 1DV The  0 9  e1   45       5  e2   57   e   45  10   3    16   e4   51   b0       e5 10   65  b1        20    e6   88   b2   e    11   7   44   e8   87  20       15   e9   89    e   59  - 15 grade   10    264 • Y=BX+E • Simple way of representing as many IVs as you like Y = b0x0 + b1x1 + b2x2 + b3x3 + b4x4 + b5x5 + e  x01   x02 x11 x21 x31 x41 x12 x22 x32 x42  b0     b1  x51  b2   e1       x52  b3   e2  b   4 b   5 265  x01 x11 x21 x31 x41   x02 x12 x22 x32 x42  b0     b1  x51  b2   e1       x52  b3   e2  b   4 b   5  b0 x0  b1 x1  ...bk xk  e 266 Generalises to Multivariate Case • Y=BX+E • Y, B and E – Matrices, not vectors • Goes beyond this course – (Do Jacques Tacq’s course for more) – (Or read his book) 267 268 269 270 Lesson 6: More on Multiple Regression 271 Parameter Estimates • Parameter estimates (b1, b2 … bk) were standardised – Because we analysed a correlation matrix • Represent the correlation of each IV with the DV – When all other IVs are held constant 272 • Can also be unstandardised • Unstandardised represent the unit change in the DV associated with a 1 unit change in the IV – When all the other variables are held constant • Parameters have standard errors associated with them – As with one IV – Hence t-test, and associated probability can be calculated • Trickier than with one IV 273 Standard Error of Regression Coefficient • Standardised is easier 1 R 1 SEi  2 n  k 1 1  R i 2 Y – R2i is the value of R2 when all other predictors are used as predictors of that variable • Note that if R2i = 0, the equation is the same as for previous 274 Multiple R • The degree of prediction – R (or Multiple R) – No longer equal to b • R2 Might be equal to the sum of squares of B – Only if all x’s are uncorrelated 275 In Terms of Variance • Can also think of this in terms of variance explained. – Each IV explains some variance in the DV – The IVs share some of their variance • Can’t share the same variance twice 276 Variance in Y accounted for by x1 rx1y2 = 0.36 The total variance of Y =1 Variance in Y accounted for by x2 rx2y2 = 0.36 277 • In this model – R2 = ryx12 + ryx22 – R2 = 0.36 + 0.36 = 0.72 – R = 0.72 = 0.85 • But – If x1 and x2 are correlated – No longer the case 278 Variance in Y accounted for by x1 rx1y2 = 0.36 The total variance of Y =1 Variance shared between x1 and x2 (not equal to rx1x2) Variance in Y accounted for by x2 rx2y2 = 0.36 279 • So – We can no longer sum the r2 – Need to sum them, and subtract the shared variance – i.e. the correlation • But – It’s not the correlation between them – It’s the correlation between them as a proportion of the variance of Y • Two different ways 280 • Based on estimates 2 R  b1ryx1  b2ryx2 • If rx1x2 = 0 – rxy = bx1 – Equivalent to ryx12 + ryx22 281 • Based on correlations 2 R  2 yx1 r 2 yx2 r  2ryx1 ryx2 rx1 x2 2 x1 x2 1r • rx1x2 = 0 – Equivalent to ryx12 + ryx22 282 • Can also be calculated using methods we have seen – Based on PRE – Based on correlation with prediction • Same procedure with >2 IVs 283 Adjusted R2 • R2 is an overestimate of population value of R2 – Any x will not correlate 0 with Y – Any variation away from 0 increases R – Variation from 0 more pronounced with lower N • Need to correct R2 – Adjusted R2 284 • Calculation of Adj. R2 N 1 Adj. R  1  (1  R ) N  k 1 2 2 • 1 – R2 – Proportion of unexplained variance – We multiple this by an adjustment • More variables – greater adjustment • More people – less adjustment 285 Shrunken R2 • Some authors treat shrunken and adjusted R2 as the same thing – Others don’t 286 N 1 N  k 1 N  20, k  3 20  1 19   1.1875 20  3  1 16 N  10, k  8 N  10, k  3 10  1 9  9 10  8  1 1 10  1 9   1.5 10  3  1 6 287 Extra Bits • Some stranger things that can happen – Counter-intuitive 288 Suppressor variables • Can be hard to understand – Very counter-intuitive • Definition – An independent variable which increases the size of the parameters associated with other independent variables above the size of their correlations 289 • An example (based on Horst, 1941) – Success of trainee pilots – Mechanical ability (x1), verbal ability (x2), success (y) • Correlation matrix Mech Mech Verb Success 1 0.5 0.3 Verb 0.5 1 0 Success 0.3 0 1 290 – Mechanical ability correlates 0.3 with success – Verbal ability correlates 0.0 with success – What will the parameter estimates be? – (Don’t look ahead until you have had a guess) 291 • Mechanical ability – b = 0.4 – Larger than r! • Verbal ability – b = -0.2 – Smaller than r!! • So what is happening? – You need verbal ability to do the test – Not related to mechanical ability • Measure of mechanical ability is contaminated by verbal ability 292 • High mech, low verbal – High mech • This is positive – Low verbal • Negative, because we are talking about standardised scores • Your mech is really high – you did well on the mechanical test, without being good at the words • High mech, high verbal – Well, you had a head start on mech, because of verbal, and need to be brought down a bit 293 Another suppressor? x1 x2 y x1 1 0.5 0.3 x2 0.5 1 0.2 y 0.3 0.2 1 b1 = b2 = 294 Another suppressor? x1 x2 y x1 1 0.5 0.3 x2 0.5 1 0.2 y 0.3 0.2 1 b1 =0.26 b2 = -0.06 295 And another? x1 x2 y x1 1 0.5 0.3 x2 0.5 1 -0.2 y 0.3 -0.2 1 b1 = b2 = 296 And another? x1 x2 y x1 1 0.5 0.3 x2 0.5 1 -0.2 y 0.3 -0.2 1 b1 = 0.53 b2 = -0.47 297 One more? x1 x2 y x1 1 -0.5 0.3 x2 -0.5 1 0.2 y 0.3 0.2 1 b1 = b2 = 298 One more? x1 x2 y x1 1 -0.5 0.3 x2 -0.5 1 0.2 y 0.3 0.2 1 b1 = 0.53 b2 = 0.47 299 • Suppression happens when two opposing forces are happening together – And have opposite effects • Don’t throw away your IVs, – Just because they are uncorrelated with the DV • Be careful in interpretation of regression estimates – Really need the correlations too, to interpret what is going on – Cannot compare between studies with different IVs 300 Standardised Estimates > 1 • Correlations are bounded -1.00 ≤ r ≤ +1.00 – We think of standardised regression estimates as being similarly bounded • But they are not – Can go >1.00, <-1.00 – R cannot, because that is a proportion of variance 301 • Three measures of ability – Mechanical ability, verbal ability 1, verbal ability 2 – Score on science exam Mech Mech Verbal1 Verbal2 Scores 1 0.1 0.1 0.6 Verbal1 0.1 1 0.9 0.6 Verbal2 0.1 0.9 1 0.3 Scores 0.6 0.6 0.3 1 –Before reading on, what are the parameter estimates? 302 Mech Verbal1 Verbal2 0.56 1.71 -1.29 • Mechanical – About where we expect • Verbal 1 – Very high • Verbal 2 – Very low 303 • What is going on – It’s a suppressor again – An independent variable which increases the size of the parameters associated with other independent variables above the size of their correlations • Verbal 1 and verbal 2 are correlated so highly – They need to cancel each other out 304 Variable Selection • What are the appropriate independent variables to use in a model? – Depends what you are trying to do • Multiple regression has two separate uses – Prediction – Explanation 305 • Prediction – What will happen in the future? – Emphasis on practical application – Variables selected (more) empirically – Value free • Explanation – Why did something happen? – Emphasis on understanding phenomena – Variables selected theoretically – Not value free 306 • Visiting the doctor – Precedes suicide attempts – Predicts suicide • Does not explain suicide • More on causality later on … • Which are appropriate variables – To collect data on? – To include in analysis? – Decision needs to be based on theoretical knowledge of the behaviour of those variables – Statistical analysis of those variables (later) • Unless you didn’t collect the data – Common sense (not a useful thing to say) 307 Variable Entry Techniques • Entry-wise – All variables entered simultaneously • Hierarchical – Variables entered in a predetermined order • Stepwise – Variables entered according to change in R2 – Actually a family of techniques 308 • Entrywise – All variables entered simultaneously – All treated equally • Hierarchical – Entered in a theoretically determined order – Change in R2 is assessed, and tested for significance – e.g. sex and age • Should not be treated equally with other variables • Sex and age MUST be first – Confused with hierarchical linear modelling 309 • Stepwise – Variables entered empirically – Variable which increases R2 the most goes first • Then the next … – Variables which have no effect can be removed from the equation • Example – IVs: Sex, age, extroversion, – DV: Car – how long someone spends looking after their car 310 • Correlation Matrix SEX SEX AGE EXTRO CAR AGE 1.00 -0.05 0.40 0.66 -0.05 1.00 0.40 0.23 EXTRO CAR 0.40 0.66 0.40 0.23 1.00 0.67 0.67 1.00 311 • Entrywise analysis – r2 = 0.64 SEX AGE EXTRO b 0.49 0.08 0.44 p <0.01 0.46 <0.01 312 • Stepwise Analysis – Data determines the order – Model 1: Extroversion, R2 = 0.450 – Model 2: Extroversion + Sex, R2 = 0.633 EXTRO SEX b 0.48 0.47 p <0.01 <0.01 313 • Hierarchical analysis – Theory determines the order – Model 1: Sex + Age, R2 = 0.510 – Model 2: S, A + E, R2 = 0.638 – Change in R2 = 0.128, p = 0.001 2 SEX AGE EXTRO 0.49 0.08 0.44 <0.01 0.46 <0.01 314 • Which is the best model? – Entrywise – OK – Stepwise – excluded age • Did have a (small) effect – Hierarchical • The change in R2 gives the best estimate of the importance of extroversion • Other problems with stepwise – F and df are wrong (cheats with df) – Unstable results • Small changes (sampling variance) – large differences in models 315 – Uses a lot of paper – Don’t use a stepwise procedure to pack your suitcase 316 Is Stepwise Always Evil? • Yes • All right, no • Research goal is predictive (technological) – Not explanatory (scientific) – What happens, not why • N is large – 40 people per predictor, Cohen, Cohen, Aiken, West (2003) • Cross validation takes place 317 A quick note on R2 R2 is sometimes regarded as the ‘fit’ of a regression model – Bad idea • If good fit is required – maximise R2 – Leads to entering variables which do not make theoretical sense 318 Critique of Multiple Regression • Goertzel (2002) – “Myths of murder and multiple regression” – Skeptical Inquirer (Paper B1) • Econometrics and regression are ‘junk science’ – Multiple regression models (in US) – Used to guide social policy 319 More Guns, Less Crime – (controlling for other factors) • Lott and Mustard: A 1% increase in gun ownership – 3.3% decrease in murder rates • But: – More guns in rural Southern US – More crime in urban North (crack cocaine epidemic at time of data) 320 Executions Cut Crime • No difference between crimes in states in US with or without death penalty • Ehrlich (1975) controlled all variables that effect crime rates – Death penalty had effect in reducing crime rate • No statistical way to decide who’s right 321 Legalised Abortion • Donohue and Levitt (1999) – Legalised abortion in 1970’s cut crime in 1990’s • Lott and Whitley (2001) – “Legalising abortion decreased murder rates by … 0.5 to 7 per cent.” • It’s impossible to model these data – Controlling for other historical events – Crack cocaine (again) 322 Another Critique • Berk (2003) – Regression analysis: a constructive critique (Sage) • Three cheers for regression – As a descriptive technique • Two cheers for regression – As an inferential technique • One cheer for regression – As a causal analysis 323 Is Regression Useless? • Do regression carefully – Don’t go beyond data which you have a strong theoretical understanding of • Validate models – Where possible, validate predictive power of models in other areas, times, groups • Particularly important with stepwise 324 Lesson 7: Categorical Independent Variables 325 Introduction 326 Introduction • So far, just looked at continuous independent variables • Also possible to use categorical (nominal, qualitative) independent variables – e.g. Sex; Job; Religion; Region; Type (of anything) • Usually analysed with t-test/ANOVA 327 Historical Note • But these (t-test/ANOVA) are special cases of regression analysis – Aspects of General Linear Models (GLMs) • So why treat them differently? – Fisher’s fault – Computers’ fault • Regression, as we have seen, is computationally difficult – Matrix inversion and multiplication – Unfeasible, without a computer 328 • In the special cases where: • You have one categorical IV • Your IVs are uncorrelated – It is much easier to do it by partitioning of sums of squares • These cases – Very rare in ‘applied’ research – Very common in ‘experimental’ research • Fisher worked at Rothamsted agricultural research station • Never have problems manipulating wheat, pigs, cabbages, etc 329 • In psychology – Led to a split between ‘experimental’ psychologists and ‘correlational’ psychologists – Experimental psychologists (until recently) would not think in terms of continuous variables • Still (too) common to dichotomise a variable – Too difficult to analyse it properly – Equivalent to discarding 1/3 of your data 330 The Approach 331 The Approach • Recode the nominal variable – Into one, or more, variables to represent that variable • Names are slightly confusing – Some texts talk of ‘dummy coding’ to refer to all of these techniques – Some (most) refer to ‘dummy coding’ to refer to one of them – Most have more than one name 332 • If a variable has g possible categories it is represented by g-1 variables • Simplest case: – Smokes: Yes or No – Variable 1 represents ‘Yes’ – Variable 2 is redundant • If it isn’t yes, it’s no 333 The Techniques 334 • We will examine two coding schemes – Dummy coding • For two groups • For >2 groups – Effect coding • For >2 groups • Look at analysis of change – Equivalent to ANCOVA – Pretest-posttest designs 335 Dummy Coding – 2 Groups • Also called simple coding by SPSS • A categorical variable with two groups • One group chosen as a reference group – The other group is represented in a variable • e.g. 2 groups: Experimental (Group 1) and Control (Group 0) – Control is the reference group – Dummy variable represents experimental group • Call this variable ‘group1’ 336 • For variable ‘group1’ – 1 = ‘Yes’, 2=‘No’ Original Category Exp Con New Variable 1 0 337 • Some data • Group is x, score is y Control Group Experiment 1 Experiment 2 Experiment 3 Experimental Group 10 10 10 20 10 30 338 • Control Group = 0 – Intercept = Score on Y when x = 0 – Intercept = mean of control group • Experimental Group = 1 – b = change in Y when x increases 1 unit – b = difference between experimental group and control group 339 35 30 Gradient of slope 25 represents difference between means 20 15 10 5 0 Control Group Experiment 1 Experimental Group Experiment 2 Experiment 3 340 Dummy Coding – 3+ Groups • With three groups the approach is the similar • g = 3, therefore g-1 = 2 variables needed • 3 Groups – Control – Experimental Group 1 – Experimental Group 2 341 Original Category Con Gp1 Gp2 Gp1 Gp2 0 1 0 0 0 1 • Recoded into two variables – Note – do not need a 3rd variable • If we are not in group 1 or group 2 MUST be in control group • 3rd variable would add no information • (What would happen to determinant?) 342 • F and associated p – Tests H0 that g1  g2  g3 • b1 and b2 and associated p-values – Test difference between each experimental group and the control group • To test difference between experimental groups – Need to rerun analysis 343 • One more complication – Have now run multiple comparisons – Increases a – i.e. probability of type I error • Need to correct for this – Bonferroni correction – Multiply given p-values by two/three (depending how many comparisons were made) 344 Effect Coding • Usually used for 3+ groups • Compares each group (except the reference group) to the mean of all groups – Dummy coding compares each group to the reference group. • Example with 5 groups – 1 group selected as reference group • Group 5 345 • Each group (except reference) has a variable – 1 if the individual is in that group – 0 if not – -1 if in reference group group 1 2 3 4 5 group_1 group_2 group_3 group_4 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 -1 -1 -1 -1 346 Examples • Dummy coding and Effect Coding • Group 1 chosen as reference group each time • Data Group Mean SD 1 52.40 4.60 2 56.30 5.70 3 60.10 5.00 Total 56.27 5.88 347 • Dummy Group dummy2 dummy3 1 2 3 0 1 0 0 0 1 Group Effect2 effect3 1 2 3 -1 1 0 -1 0 1 • Effect 348 Dummy R=0.543, F=5.7, df=2, 27, p=0.009 b0 = 52.4, b1 = 3.9, p=0.100 b2 = 7.7, p=0.002 Effect R=0.543, F=5.7, df=2, 27, p=0.009 b0 = 56.27, b1 = 0.03, p=0.980 b2 = 3.8, p=0.007 b0  g1 b0  G b1  g2  g1 b1  g2  G b2  g3  g1 b2  g3  G 349 In SPSS • SPSS provides two equivalent procedures for regression – Regression (which we have been using) – GLM (which we haven’t) • GLM will: – Automatically code categorical variables – Automatically calculate interaction terms • GLM won’t: – Give standardised effects – Give hierarchical R2 p-values – Allow you to not understand 350 ANCOVA and Regression 351 • Test – (Which is a trick; but it’s designed to make you think about it) • Use employee data.sav – Compare the pay rise (difference between salbegin and salary) – For ethnic minority and non-minority staff • What do you find? 352 ANCOVA and Regression • Dummy coding approach has one special use – In ANCOVA, for the analysis of change • Pre-test post-test experimental design – Control group and (one or more) experimental groups – Tempting to use difference score + t-test / mixed design ANOVA – Inappropriate 353 • Salivary cortisol levels – Used as a measure of stress – Not absolute level, but change in level over day may be interesting • Test at: 9.00am, 9.00pm • Two groups – High stress group (cancer biopsy) • Group 1 – Low stress group (no biopsy) • Group 0 354 High Stress Low Stress AM 20.1 22.3 PM 6.8 11.8 Diff 13.3 10.5 • Correlation of AM and PM = 0.493 (p=0.008) • Has there been a significant difference in the rate of change of salivary cortisol? – 3 different approaches 355 • Approach 1 – find the differences, do a t-test – t = 1.31, df=26, p=0.203 • Approach 2 – mixed ANOVA, look for interaction effect – F = 1.71, df = 1, 26, p = 0.203 – F = t2 • Approach 3 – regression (ANCOVA) based approach 356 – IVs: AM and group – DV: PM – b1 (group) = 3.59, standardised b1=0.432, p = 0.01 • Why is the regression approach better? – The other two approaches took the difference – Assumes that r = 1.00 – Any difference from r = 1.00 and you add error variance • Subtracting error is the same as adding error 357 • Using regression – Ensures that all the variance that is subtracted is true – Reduces the error variance • Two effects – Adjusts the means • Compensates for differences between groups – Removes error variance 358 In SPSS • SPSS automates all of this – But you have to understand it, to know what it is doing • Use Analyse, GLM, Univariate ANOVA 359 Outcome here Categorical predictors here Continuous predictors here Click options 360 Select parameter estimaters 361 More on Change • If difference score is correlated with either pre-test or post-test – Subtraction fails to remove the difference between the scores – If two scores are uncorrelated • Difference will be correlated with both • Failure to control – Equal SDs, r = 0 • Correlation of change and pre-score =0.707 362 Even More on Change • A topic of surprising complexity – What I said about difference scores isn’t always true • Lord’s paradox – it depends on the precise question you want to answer – Collins and Horn (1993). Best methods for the analysis of change – Collins and Sayer (2001). New methods for the analysis of change. 363 Lesson 8: Assumptions in Regression Analysis 364 The Assumptions 1. The distribution of residuals is normal (at each value of the dependent variable). 2. The variance of the residuals for every set of values for the independent variable is equal. • violation is called heteroscedasticity. 3. The error term is additive • no interactions. 4. At every value of the dependent variable the expected (mean) value of the residuals is zero • No non-linear relationships 365 5. The expected correlation between residuals, for any two cases, is 0. • The independence assumption (lack of autocorrelation) 6. All independent variables are uncorrelated with the error term. 7. No independent variables are a perfect linear function of other independent variables (no perfect multicollinearity) 8. The mean of the error term is zero. 366 What are we going to do … • Deal with some of these assumptions in some detail • Deal with others in passing only – look at them again later on 367 Assumption 1: The Distribution of Residuals is Normal at Every Value of the Dependent Variable 368 Look at Normal Distributions • A normal distribution – symmetrical, bell-shaped (so they say) 369 What can go wrong? • Skew – non-symmetricality – one tail longer than the other • Kurtosis – too flat or too peaked – kurtosed • Outliers – Individual cases which are far from the distribution 370 Effects on the Mean • Skew – biases the mean, in direction of skew • Kurtosis – mean not biased – standard deviation is – and hence standard errors, and significance tests 371 Examining Univariate Distributions • • • • Histograms Boxplots P-P plots Calculation based methods 372 Histograms 30 A and B 30 20 20 10 10 0 0 373 • C and D 40 14 12 30 10 8 20 6 4 10 2 0 0 374 •E&F 20 10 0 375 Histograms can be tricky …. 7 6 6 5 5 4 4 3 3 2 2 1 1 0 0 0 7 7 6 6 6 5 5 6 5 4 3 2 1 5 4 4 4 3 3 2 2 1 1 0 0 3 2 1 0 376 Boxplots 377 P-P Plots •A&B 1.00 1.00 .75 .75 .50 .50 .25 .25 0.00 0.00 .25 .50 .75 1.00 0.00 0.00 .25 .50 .75 1.00 378 •C&D 1.00 1.00 .75 .75 .50 .50 .25 .25 0.00 0.00 .25 .50 .75 1.00 0.00 0.00 .25 .50 .75 1.00 379 •E&F 1.00 1.00 .75 .75 .50 .50 .25 .25 0.00 0.00 .25 .50 .75 1.00 0.00 0.00 .25 .50 .75 1.00 380 Calculation Based • Skew and Kurtosis statistics • Outlier detection statistics 381 Skew and Kurtosis Statistics • Normal distribution – skew = 0 – kurtosis = 0 • Two methods for calculation – Fisher’s and Pearson’s – Very similar answers • Associated standard error – can be used for significance of departure from normality – not actually very useful • Never normal above N = 400 382 Skewness SE Skew Kurtosis SE Kurt A B C D E F -0.12 0.271 0.454 0.117 2.106 0.171 0.172 0.172 0.172 0.172 0.172 0.172 -0.084 0.265 1.885 -1.081 5.75 -0.21 0.342 0.342 0.342 0.342 0.342 0.342 383 Outlier Detection • Calculate distance from mean – z-score (number of standard deviations) – deleted z-score • that case biased the mean, so remove it – Look up expected distance from mean • 1% 3+ SDs • Calculate influence – how much effect did that case have on the mean? 384 Non-Normality in Regression 385 Effects on OLS Estimates • The mean is an OLS estimate • The regression line is an OLS estimate • Lack of normality – biases the position of the regression slope – makes the standard errors wrong • probability values attached to statistical significance wrong 386 Checks on Normality • Check residuals are normally distributed – SPSS will draw histogram and p-p plot of residuals • Use regression diagnostics – Lots of them – Most aren’t very interesting 387 Regression Diagnostics • Residuals – standardised, unstandardised, studentised, deleted, studentised-deleted – look for cases > |3| (?) • Influence statistics – Look for the effect a case has – If we remove that case, do we get a different answer? – DFBeta, Standardised DFBeta • changes in b 388 – DfFit, Standardised DfFit • change in predicted value – Covariance ratio • Ratio of the determinants of the covariance matrices, with and without the case • Distances – measures of ‘distance’ from the centroid – some include IV, some don’t 389 More on Residuals • Residuals are trickier than you might have imagined • Raw residuals – OK • Standardised residuals – Residuals divided by SD se  e n  k 1 2 390 Leverage • But – That SD is wrong – Variance of the residuals is not equal • Those further from the centroid on the predictors have higher variance • Need a measure of this • Distance from the centroid is leverage, or h (or sometimes hii) • One predictor – Easy 391  xi  x  1 hi   2 n ( x  x ) 2 • Minimum hi is 1/n, the maximum is 1 • Except – SPSS uses standardised leverage - h* • It doesn’t tell you this, it just uses it 392 1 hi  hi  n 2  xi  x  * hi  2 ( x  x ) * • Minimum 0, maximum (N-1/N) 393 • Multiple predictors – Calculate the hat matrix (H) – Leverage values are the diagonals of this matrix 1 H  X(X' X) X' – Where X is the augmented matrix of predictors (i.e. matrix that includes the constant) – Hence leverage hii – element ii of H 394 • Example of calculation of hat matrix 1      1 15   1 15   1 15   1 15   0.318 0.273             1 20   1 20   1 20   1 20   0.273 0.236  H        ... ...    ... ...   ... ...   ... ...             0.318   1 65   1 65   1 65   1 65     395 Standardised / Studentised • Now we can calculate the standardised residuals – SPSS calls them studentised residuals – Also called internally studentised residuals ei ei  se 1  hi 396 Deleted Studentised Residuals • Studentised residuals do not have a known distribution – Cannot use them for inference • Deleted studentised residuals – Externally studentised residuals – Jackknifed residuals • Distributed as t • With df = N – k – 1 397 Testing Significance • We can calculate the probability of a residual – Is it sampled from the same population • BUT – Massive type I error rate – Bonferroni correct it • Multiply p value by N 398 Bivariate Normality • We didn’t just say “residuals normally distributed” • We said “at every value of the dependent variables” • Two variables can be normally distributed – univariate, – but not bivariate 399 • Couple’s IQs – male and female FEMALE MALE 8 6 5 6 4 4 3 2 Frequency 2 0 60.0 70.0 80.0 90.0 100.0 110.0 120.0 130.0 1 0 140.0 60.0 70.0 80.0 90.0 100.0 110.0 120.0 130.0 140.0 –Seem reasonably normal 400 • But wait!! 160 140 120 100 80 MALE 60 40 40 60 80 100 120 140 160 FEMALE 401 • When we look at bivariate normality – not normal – there is an outlier • So plot X against Y • OK for bivariate – but – may be a multivariate outlier – Need to draw graph in 3+ dimensions – can’t draw a graph in 3 dimensions • But we can look at the residuals instead … 402 • IQ histogram of residuals 12 10 8 6 4 2 0 403 Multivariate Outliers … • Will be explored later in the exercises • So we move on … 404 What to do about NonNormality • Skew and Kurtosis – Skew – much easier to deal with – Kurtosis – less serious anyway • Transform data – removes skew – positive skew – log transform – negative skew - square 405 Transformation • May need to transform IV and/or DV – More often DV • time, income, symptoms (e.g. depression) all positively skewed – can cause non-linear effects (more later) if only one is transformed – alters interpretation of unstandardised parameter – May alter meaning of variable – May add / remove non-linear and moderator effects 406 • Change measures – increase sensitivity at ranges • avoiding floor and ceiling effects • Outliers – Can be tricky – Why did the outlier occur? • Error? Delete them. • Weird person? Probably delete them • Normal person? Tricky. 407 – You are trying to model a process • is the data point ‘outside’ the process • e.g. lottery winners, when looking at salary • yawn, when looking at reaction time – Which is better? • A good model, which explains 99% of your data? • A poor model, which explains all of it • Pedhazur and Schmelkin (1991) – analyse the data twice 408 • We will spend much less time on the other 6 assumptions • Can do exercise 8.1. 409 Assumption 2: The variance of the residuals for every set of values for the independent variable is equal. 410 Heteroscedasticity • This assumption is a about heteroscedasticity of the residuals – Hetero=different – Scedastic = scattered • We don’t want heteroscedasticity – we want our data to be homoscedastic • Draw a scatterplot to investigate 411 160 140 120 100 80 MALE 60 40 40 60 FEMALE 80 100 120 140 160 412 • Only works with one IV – need every combination of IVs • Easy to get – use predicted values – use residuals there • Plot predicted values against residuals – or – or – or – or standardised residuals deleted residuals standardised deleted residuals studentised residuals • A bit like turning the scatterplot on its side 413 Good – no heteroscedasticity Predicted Value 414 Bad – heteroscedasticity Predicted Value 415 Testing Heteroscedasticity • White’s test – – 1. 2. 3. 4. Not automatic in SPSS (is in SAS) Luckily, not hard to do Do regression, save residuals. Square residuals Square IVs Calculate interactions of IVs – e.g. x1•x2, x1•x3, x2 • x3 416 5. Run regression using – squared residuals as DV – IVs, squared IVs, and interactions as IVs 6. Test statistic = N x R2 – Distributed as c2 – Df = k (for second regression) • Use education and salbegin to predict salary (employee data.sav) – R2 = 0.113, N=474, c2 = 53.5, df=5, p < 0.0001 417 Plot of Pred and Res 8 6 4 2 0 -2 -4 -2 0 2 4 6 8 Regression Standardized Predicted Value 418 Magnitude of Heteroscedasticity • Chop data into “slices” – 5 slices, based on X (or predicted score) • Done in SPSS – Calculate variance of each slice – Check ratio of smallest to largest – Less than 10:1 • OK 419 The Visual Bander • New in SPSS 12 420 1 • Variances of the 5 groups .219 2 .336 3 .757 4 .751 5 3.119 • We have a problem – 3 / 0.2 ~= 15 421 Dealing with Heteroscedasticity • Use Huber-White estimates – Very easy in Stata – Fiddly in SPSS – bit of a hack • Use Complex samples 1. Create a new variable where all cases are equal to 1, call it const 2. Use Complex Samples, Prepare for Analysis 3. Create a plan file 422 4. 5. 6. 7. Sample weight is const Finish Use Complex Samples, GLM Use plan file created, and set up model as in GLM (More on complex samples later) In Stata, do regression as normal, and click “robust”. 423 Heteroscedasticity – Implications and Meanings Implications • What happens as a result of heteroscedasticity? – Parameter estimates are correct • not biased – Standard errors (hence p-values) are incorrect 424 However … • If there is no skew in predicted scores – P-values a tiny bit wrong • If skewed, – P-values very wrong • Can do exercise 425 Meaning • What is heteroscedasticity trying to tell us? – Our model is wrong – it is misspecified – Something important is happening that we have not accounted for • e.g. amount of money given to charity (given) – depends on: • earnings • degree of importance person assigns to the charity (import) 426 • Do the regression analysis – R2 = 0.60, F=31.4, df=2, 37, p < 0.001 • seems quite good – b0 = 0.24, p=0.97 – b1 = 0.71, p < 0.001 – b2 = 0.23, p = 0.031 • White’s test – c2 = 18.6, df=5, p=0.002 • The plot of predicted values against residuals … 427 • Plot shows heteroscedastic relationship 428 • Which means … – the effects of the variables are not additive – If you think that what a charity does is important • you might give more money • how much more depends on how much money you have 429 70 60 50 40 30 GIVEN Earnings 20 High 10 Low 4 6 8 10 12 14 16 IMPORT 430 • One more thing about heteroscedasticity – it is the equivalent of homogeneity of variance in ANOVA/t-tests 431 Assumption 3: The Error Term is Additive 432 Additivity • What heteroscedasticity shows you – effects of variables need to be additive • Heteroscedasticity doesn’t always show it to you – can test for it, but hard work – (same as homogeneity of covariance assumption in ANCOVA) • Have to know it from your theory • A specification error 433 Additivity and Theory • Two IVs – Alcohol has sedative effect • A bit makes you a bit tired • A lot makes you very tired – Some painkillers have sedative effect • A bit makes you a bit tired • A lot makes you very tired – A bit of alcohol and a bit of painkiller doesn’t make you very tired – Effects multiply together, don’t add together 434 • If you don’t test for it – It’s very hard to know that it will happen • So many possible non-additive effects – Cannot test for all of them – Can test for obvious • In medicine – Choose to test for salient non-additive effects – e.g. sex, race 435 Assumption 4: At every value of the dependent variable the expected (mean) value of the residuals is zero 436 Linearity • Relationships between variables should be linear – best represented by a straight line • Not a very common problem in social sciences – except economics – measures are not sufficiently accurate to make a difference • R2 too low • unlike, say, physics 437 Fuel • Relationship between speed of travel and fuel used Speed 438 • R2 = 0.938 – looks pretty good – know speed, make a good prediction of fuel • BUT – look at the chart – if we know speed we can make a perfect prediction of fuel used – R2 should be 1.00 439 Detecting Non-Linearity • Residual plot – just like heteroscedasticity • Using this example – very, very obvious – usually pretty obvious 440 Residual plot 441 Linearity: A Case of Additivity • Linearity = additivity along the range of the IV • Jeremy rides his bicycle harder – Increase in speed depends on current speed – Not additive, multiplicative – MacCallum and Mar (1995). Distinguishing between moderator and quadratic effects in multiple regression. Psychological Bulletin. 442 Assumption 5: The expected correlation between residuals, for any two cases, is 0. The independence assumption (lack of autocorrelation) 443 Independence Assumption • Also: lack of autocorrelation • Tricky one – often ignored – exists for almost all tests • All cases should be independent of one another – knowing the value of one case should not tell you anything about the value of other cases 444 How is it Detected? • Can be difficult – need some clever statistics (multilevel models) • Better off avoiding situations where it arises • Residual Plots • Durbin-Watson Test 445 Residual Plots • Were data collected in time order? – If so plot ID number against the residuals – Look for any pattern • Test for linear relationship • Non-linear relationship • Heteroscedasticity 446 2 Residual 1 0 -1 -2 0 10 20 30 40 Participant Number 447 How does it arise? Two main ways • time-series analyses – When cases are time periods • weather on Tuesday and weather on Wednesday correlated • inflation 1972, inflation 1973 are correlated • clusters of cases – patients treated by three doctors – children from different classes – people assessed in groups 448 Why does it matter? • Standard errors can be wrong – therefore significance tests can be wrong • Parameter estimates can be wrong – really, really wrong – from positive to negative • An example – students do an exam (on statistics) – choose one of three questions • IV: time • DV: grade 449 •Result, with line of best fit 90 80 70 60 50 40 Grade 30 20 10 10 Time 20 30 40 50 60 70 450 • Result shows that – people who spent longer in the exam, achieve better grades • BUT … – we haven’t considered which question people answered – we might have violated the independence assumption • DV will be autocorrelated • Look again – with questions marked 451 • Now somewhat different 90 80 70 60 50 40 Question 30 Grade 3 20 2 10 10 1 20 30 40 50 60 70 Time 452 • Now, people that spent longer got lower grades – questions differed in difficulty – do a hard one, get better grade – if you can do it, you can do it quickly • Very difficult to analyse well – need multilevel models 453 Durbin Watson Test • Not well implemented in SPSS • Depends on the order of the data – Reorder the data, get a different result • Doesn’t give statistical significance of the test 454 Assumption 6: All independent variables are uncorrelated with the error term. 455 Uncorrelated with the Error Term • A curious assumption – by definition, the residuals are uncorrelated with the independent variables (try it and see, if you like) • It is about the DV – must have no effect (when the IVs have been removed) – on the DV 456 • Problem in economics – Demand increases supply – Supply increases wages – Higher wages increase demand • OLS estimates will be (badly) biased in this case – need a different estimation procedure – two-stage least squares • simultaneous equation modelling 457 Assumption 7: No independent variables are a perfect linear function of other independent variables no perfect multicollinearity 458 No Perfect Multicollinearity • IVs must not be linear functions of one another – matrix of correlations of IVs is not positive definite – cannot be inverted – analysis cannot proceed • Have seen this with – age, age start, time working – also occurs with subscale and total 459 • Large amounts of collinearity – a problem (as we shall see) sometimes – not an assumption 460 Assumption 8: The mean of the error term is zero. You will like this one. 461 Mean of the Error Term = 0 • Mean of the residuals = 0 • That is what the constant is for – if the mean of the error term deviates from zero, the constant soaks it up Y   0  1 x1   Y  (  0  3)  1 x1  (  3) - note, Greek letters because we are talking about population values 462 • Can do regression without the constant – Usually a bad idea – E.g R2 = 0.995, p < 0.001 • Looks good 463 13 12 y 11 10 9 8 7 6 7 8 9 10 11 12 13 x1 464 465 Lesson 9: Issues in Regression Analysis Things that alter the interpretation of the regression equation 466 The Four Issues • • • • Causality Sample sizes Collinearity Measurement error 467 Causality 468 What is a Cause? • Debate about definition of cause – some statistics (and philosophy) books try to avoid it completely – We are not going into depth • just going to show why it is hard • Two dimensions of cause – Ultimate versus proximal cause – Determinate versus probabilistic 469 Proximal versus Ultimate • Why am I here? – I walked here because – This is the location of the class because – Eric Tanenbaum asked me because – (I don’t know) – because I was in my office when he rang because – I am a lecturer at York because – I saw an advert in the paper because 470 – I exist because – My parents met because – My father had a job … • Proximal cause – the direct and immediate cause of something • Ultimate cause – the thing that started the process off – I fell off my bicycle because of the bump – I fell off because I was going too fast 471 Determinate versus Probabilistic Cause • Why did I fall off my bicycle? – I was going too fast – But every time I ride too fast, I don’t fall off – Probabilistic cause • Why did my tyre go flat? – A nail was stuck in my tyre – Every time a nail sticks in my tyre, the tyre goes flat – Deterministic cause 472 • Can get into trouble by mixing them together – Eating deep fried Mars Bars and doing no exercise are causes of heart disease – “My Grandad ate three deep fried Mars Bars every day, and the most exercise he ever got was when he walked to the shop next door to buy one” – (Deliberately?) confusing deterministic and probabilistic causes 473 Criteria for Causation • Association • Direction of Influence • Isolation 474 Association • Correlation does not mean causation – we all know • But – Causation does mean correlation • Need to show that two things are related – may be correlation – my be regression when controlling for third (or more) factor 475 • Relationship between price and sales – suppliers may be cunning – when people want it more • stick the price up Price Price Demand Sales 1 0.6 0 Demand 0.6 1 0.6 Sales 0 0.6 1 – So – no relationship between price and sales 476 – Until (or course) we control for demand – b1 (Price) = -0.56 – b2 (Demand) = 0.94 • But which variables do we enter? 477 Direction of Influence • Relationship between A and B – three possible processes A B A B B causes A A B C causes A & B C A causes B 478 • How do we establish the direction of influence? – Longitudinally? Barometer Drops Storm – Now if we could just get that barometer needle to stay where it is … • Where the role of theory comes in (more on this later) 479 Isolation • Isolate the dependent variable from all other influences – as experimenters try to do • Cannot do this – can statistically isolate the effect – using multiple regression 480 Role of Theory • Strong theory is crucial to making causal statements • Fisher said: to make causal statements “make your theories elaborate.” – don’t rely purely on statistical analysis • Need strong theory to guide analyses – what critics of non-experimental research don’t understand 481 • S.J. Gould – a critic – says correlate price of petrol and his age, for the last 10 years – find a correlation – Ha! (He says) that doesn’t mean there is a causal link – Of course not! (We say). • No social scientist would do that analysis without first thinking (very hard) about the possible causal relations between the variables of interest • Would control for time, prices, etc … 482 • Atkinson, et al. (1996) – relationship between college grades and number of hours worked – negative correlation – Need to control for other variables – ability, intelligence • Gould says “Most correlations are noncausal” (1982, p243) – Of course!!!! 483 I drink a lot of beer 16 causal relations 120 non-causal correlations laugh toilet jokes (about statistics) vomit karaoke curtains closed sleeping headache equations (beermat) thirsty fried breakfast no beer curry chips falling over lose keys 484 • Abelson (1995) elaborates on this – ‘method of signatures’ • A collection of correlations relating to the process – the ‘signature’ of the process • e.g. tobacco smoking and lung cancer – can we account for all of these findings with any other theory? 485 1. 2. 3. 4. 5. 6. 7. 8. The longer a person has smoked cigarettes, the greater the risk of cancer. The more cigarettes a person smokes over a given time period, the greater the risk of cancer. People who stop smoking have lower cancer rates than do those who keep smoking. Smoker’s cancers tend to occur in the lungs, and be of a particular type. Smokers have elevated rates of other diseases. People who smoke cigars or pipes, and do not usually inhale, have abnormally high rates of lip cancer. Smokers of filter-tipped cigarettes have lower cancer rates than other cigarette smokers. Non-smokers who live with smokers have elevated cancer rates. (Abelson, 1995: 183-184) 486 – In addition, should be no anomalous correlations • If smokers had more fallen arches than nonsmokers, not consistent with theory • Failure to use theory to select appropriate variables – specification error – e.g. in previous example – Predict wealth from price and sales • increase price, price increases • Increase sales, price increases 487 • Sometimes these are indicators of the process – e.g. barometer – stopping the needle won’t help – e.g. inflation? Indicator or cause? 488 No Causation without Experimentation • Blatantly untrue – I don’t doubt that the sun shining makes us warm • Why the aversion? – Pearl (2000) says problem is no mathematical operator – No one realised that you needed one – Until you build a robot 489 AI and Causality • A robot needs to make judgements about causality • Needs to have a mathematical representation of causality – Suddenly, a problem! – Doesn’t exist • Most operators are non-directional • Causality is directional 490 Sample Sizes “How many subjects does it take to run a regression analysis?” 491 Introduction • Social scientists don’t worry enough about the sample size required – “Why didn’t you get a significant result?” – “I didn’t have a large enough sample” • Not a common answer • More recently awareness of sample size is increasing – use too few – no point doing the research – use too many – waste their time 492 • Research funding bodies • Ethical review panels – both become more interested in sample size calculations • We will look at two approaches – Rules of thumb (quite quickly) – Power Analysis (more slowly) 493 Rules of Thumb • Lots of simple rules of thumb exist – 10 cases per IV – >100 cases – Green (1991) more sophisticated • To test significance of R2 – N = 50 + 8k • To test sig of slopes, N = 104 + k • Rules of thumb don’t take into account all the information that we have – Power analysis does 494 Power Analysis Introducing Power Analysis • Hypothesis test – tells us the probability of a result of that magnitude occurring, if the null hypothesis is correct (i.e. there is no effect in the population) • Doesn’t tell us – the probability of that result, if the null hypothesis is false 495 • According to Cohen (1982) all null hypotheses are false – everything that might have an effect, does have an effect • it is just that the effect is often very tiny 496 Type I Errors • Type I error is false rejection of H0 • Probability of making a type I error – a – the significance value cut-off • usually 0.05 (by convention) • Always this value • Not affected by – sample size – type of test 497 Type II errors • Type II error is false acceptance of the null hypothesis – Much, much trickier • We think we have some idea – we almost certainly don’t • Example – I do an experiment (random sampling, all assumptions perfectly satisfied) – I find p = 0.05 498 – You repeat the experiment exactly • different random sample from same population – What is probability you will find p < 0.05? – ……………… – Another experiment, I find p = 0.01 – Probability you find p < 0.05? – ……………… • Very hard to work out – not intuitive – need to understand non-central sampling distributions (more in a minute) 499 • Probability of type II error = beta () – same as population regression parameter (to be confusing) • Power = 1 – Beta – Probability of getting a significant result 500 State of the World Research Findings H0 True (no effect to be found) H0 false (effect to be found) H0 true (we find no effect – p > 0.05)  Type II error p= power = 1 -  H0 false (we find an effect – p < 0.05) Type I error p=a  501 • Four parameters in power analysis – a – prob. of Type I error –  – prob. of Type II error (power = 1 – ) – Effect size – size of effect in population –N • Know any three, can calculate the fourth – Look at them one at a time 502 • a Probability of Type I error – Usually set to 0.05 – Somewhat arbitrary • sometimes adjusted because of circumstances – rarely because of power analysis – May want to adjust it, based on power analysis 503 •  – Probability of type II error – Power (probability of finding a result) =1– – Standard is 80% • Some argue for 90% – Implication that Type I error is 4 times more serious than type II error • adjust ratio with compromise power analysis 504 • Effect size in the population – Most problematic to determine – Three ways 1. What effect size would be useful to find? • R2 = 0.01 - no use (probably) 2. Base it on previous research – what have other people found? 3. Use Cohen’s conventions – small R2 = 0.02 – medium R2 = 0.13 – large R2 = 0.26 505 – Effect size usually measured as f2 – For R2 2 R f  2 1 R 2 506 – For (standardised) slopes 2 sri f  2 1 R 2 – Where sr2 is the contribution to the variance accounted for by the variable of interest – i.e. sr2 = R2 (with variable) – R2 (without) • change in R2 in hierarchical regression 507 • N – the sample size – usually use other three parameters to determine this – sometimes adjust other parameters (a) based on this – e.g. You can have 50 participants. No more. 508 Doing power analysis • With power analysis program – SamplePower, GPower, Nquery • With SPSS MANOVA – using non-central distribution functions – Uses MANOVA syntax • Relies on the fact you can do anything with MANOVA • Paper B4 509 Underpowered Studies • Research in the social sciences is often underpowered – Why? – See Paper B11 – “the persistence of underpowered studies” 510 Extra Reading • Power traditionally focuses on p values – What about CIs? – Paper B8 – “Obtaining regression coefficients that are accurate, not simply significant” 511 Collinearity 512 Collinearity as Issue and Assumption • Collinearity (multicollinearity) – the extent to which the independent variables are (multiply) correlated • If R2 for any IV, using other IVs = 1.00 – perfect collinearity – variable is linear sum of other variables – regression will not proceed – (SPSS will arbitrarily throw out a variable) 513 • R2 < 1.00, but high – other problems may arise • Four things to look at in collinearity – meaning – implications – detection – actions 514 Meaning of Collinearity • Literally ‘co-linearity’ – lying along the same line • Perfect collinearity – when some IVs predict another – Total = S1 + S2 + S3 + S4 – S1 = Total – (S2 + S3 + S4) – rare 515 • Less than perfect – when some IVs are close to predicting – correlations between IVs are high (usually, but not always) 516 Implications • Effects the stability of the parameter estimates – and so the standard errors of the parameter estimates – and so the significance • Because – shared variance, which the regression procedure doesn’t know where to put 517 • Red cars have more accidents than other coloured cars – because of the effect of being in a red car? – because of the kind of person that drives a red car? • we don’t know – No way to distinguish between these three: Accidents = 1 x colour + 0 x person Accidents = 0 x colour + 1 x person Accidents = 0.5 x colour + 0.5 x person 518 • Sex differences – due to genetics? – due to upbringing? – (almost) perfect collinearity • statistically impossible to tell 519 • When collinearity is less than perfect – increases variability of estimates between samples – estimates are unstable – reflected in the variances, and hence standard errors 520 Detecting Collinearity • Look at the parameter estimates – large standardised parameter estimates (>0.3?), which are not significant • be suspicious • Run a series of regressions – each IV as DV – all other IVs as IVs • for each IV 521 • Sounds like hard work? – SPSS does it for us! • Ask for collinearity diagnostics – Tolerance – calculated for every IV Tolerance  1-R 2 – Variance Inflation Factor • sq. root of amount s.e. has been increased 1 VIF  Tolerance 522 Actions What you can do about collinearity “no quick fix” (Fox, 1991) 1. Get new data • • • avoids the problem address the question in a different way e.g. find people who have been raised as the ‘wrong’ gender • • exist, but rare Not a very useful suggestion 523 2. Collect more data • • • not different data, more data collinearity increases standard error (se) se decreases as N increases • get a bigger N 3. Remove / Combine variables • • • If an IV correlates highly with other IVs Not telling us much new If you have two (or more) IVs which are very similar • e.g. 2 measures of depression, socioeconomic status, achievement, etc 524 • • sum them, average them, remove one Many measures • use principal components analysis to reduce them 3. Use stepwise regression (or some flavour of) • • See previous comments Can be useful in theoretical vacuum 4. Ridge regression • • not very useful behaves weirdly 525 Measurement Error 526 What is Measurement Error • In social science, it is unlikely that we measure any variable perfectly – measurement error represents this imperfection • We assume that we have a true score – T • A measure of that score –x 527 x T e • just like a regression equation – standardise the parameters – T is the reliability • the amount of variance in x which comes from T • but, like a regression equation – assume that e is random and has mean of zero – more on that later 528 Simple Effects of Measurement Error • Lowers the measured correlation – between two variables • Real correlation – true scores (x* and y*) • Measured correlation – measured scores (x and y) 529 True correlation of x and y rx*y* x* e y* Reliability of x rxx Reliability of y ryy x y Measured correlation of x and y rxy e 530 • Attenuation of correlation rxy  rx * y *  rxx ryy • Attenuation corrected correlation rx * y *  rxy rxx ryy 531 • Example rxx  0.7 ryy  0.8 rxy  0.3 rx* y*  rx* y* rxy rxx ryy 0.3   0.40 0.7  0.8 532 Complex Effects of Measurement Error • Really horribly complex • Measurement error reduces correlations – reduces estimate of  – reducing one estimate • increases others – because of effects of control – combined with effects of suppressor variables – exercise to examine this 533 Dealing with Measurement Error • Attenuation correction – very dangerous – not recommended • Avoid in the first place – use reliable measures – don’t discard information • don’t categorise • Age: 10-20, 21-30, 31-40 … 534 Complications • Assume measurement error is – additive – linear • Additive – e.g. weight – people may under-report / overreport at the extremes • Linear – particularly the case when using proxy variables 535 • e.g. proxy measures – Want to know effort on childcare, count number of children • 1st child is more effort than last – Want to know financial status, count income • 1st £10 much greater effect on financial status than the 1000th. 536 Lesson 10: Non-Linear Analysis in Regression 537 Introduction • Non-linear effect occurs – when the effect of one independent variable – is not consistent across the range of the IV • Assumption is violated – expected value of residuals = 0 – no longer the case 538 Some Examples 539 Skill A Learning Curve Experience 540 Performance Yerkes-Dodson Law of Arousal Arousal 541 Suicidal Enthusiastic Enthusiasm Levels over a Lesson on Regression 0 Time 3.5 542 • Learning – line changed direction once • Yerkes-Dodson – line changed direction once • Enthusiasm – line changed direction twice 543 Everything is Non-Linear • Every relationship we look at is nonlinear, for two reasons – Exam results cannot keep increasing with reading more books • Linear in the range we examine – For small departures from linearity • Cannot detect the difference • Non-parsimonious solution 544 Non-Linear Transformations 545 Bending the Line • Non-linear regression is hard – We cheat, and linearise the data • Do linear regression Transformations • We need to transform the data – rather than estimating a curved line • which would be very difficult • may not work with OLS – we can take a straight line, and bend it – or take a curved line, and straighten it • back to linear (OLS) regression 546 • We still do linear regression – Linear in the parameters – Y = b1x + b2x2 + … • Can do non-linear regression – Non-linear in the parameters – Y = b1x + b2x2 + … • Much trickier – Statistical theory either breaks down OR becomes harder 547 • Linear transformations – multiply by a constant – add a constant – change the slope and the intercept 548 y=2x y y=x + 3 y=x x 549 • Linear transformations are no use – alter the slope and intercept – don’t alter the standardised parameter estimate • Non-linear transformation – will bend the slope – quadratic transformation y = x2 – one change of direction 550 – Cubic transformation y = x2 + x3 – two changes of direction 551 Quadratic Transformation y=0 + 0.1x + 1x2 552 Square Root Transformation y=20 + -3x + 5x 553 Cubic Transformation y = 3 - 4x + 2x2 - 0.2x3 6 5 4 3 2 1 0 0 1 2 3 4 5 6 554 Logarithmic Transformation y = 1 + 0.1x + 10log(x) 555 Inverse Transformation y = 20 -10x + 8(1/x) 556 • To estimate a non-linear regression – we don’t actually estimate anything nonlinear – we transform the x-variable to a non-linear version – can estimate that straight line – represents the curve – we don’t bend the line, we stretch the space around the line, and make it flat 557 Detecting Non-linearity 558 Draw a Scatterplot • Draw a scatterplot of y plotted against x – see if it looks a bit non-linear – e.g. Anscombe’s data – e.g. Education and beginning salary • from bank data • drawn in SPSS • with line of best fit 559 • Anscombe (1973) – constructed a set of datasets – show the importance of graphs in regression/correlation • For each dataset N Mean of x Mean of y Equation of regression line sum of squares (X - mean) correlation coefficient R2 11 9 7.5 y = 3 + 0.5x 110 0.82 0.67 560 561 562 563 564 A Real Example • Starting salary and years of education – From employee data.sav 565 Expected value of error (residual) is > 0 Educational Level (years) Expected value of error (residual) is < 0 566 Use Residual Plot • Scatterplot is only good for one variable – use the residual plot (that we used for heteroscedasticity) • Good for many variables 567 • We want – points to lie in a nice straight sausage 568 • We don’t want – a nasty bent sausage 569 • Educational level and starting salary 10 8 6 4 2 0 -2 -2 -1 0 1 2 3 570 Carrying Out Non-Linear Regression 571 Linear Transformation • Linear transformation doesn’t change – interpretation of slope – standardised slope – se, t, or p of slope – R2 • Can change – effect of a transformation 572 • Actually more complex – with some transformations can add a constant with no effect (e.g. quadratic) • With others does have an effect – inverse, log • Sometimes it is necessary to add a constant – negative numbers have no square root – 0 has no log 573 Education and Salary Linear Regression • Saw previously that the assumption of expected errors = 0 was violated • Anyway … – R2 = 0.401, F=315, df = 1, 472, p < 0.001 – salbegin = -6290 + 1727  educ – Standardised • b1 (educ) = 0.633 – Both parameters make sense 574 Non-linear Effect • Compute new variable – quadratic – educ2 = educ2 • Add this variable to the equation – R2 = 0.585, p < 0.001 – salbegin = 46263 + -6542  educ + 310  educ2 • slightly curious – Standardised • b1 (educ) = -2.4 • b2 (educ2) = 3.1 – What is going on? 575 • Collinearity – is what is going on – Correlation of educ and educ2 • r = 0.990 – Regression equation becomes difficult (impossible?) to interpret • Need hierarchical regression – what is the change in R2 – is that change significant? – R2 (change) = 0.184, p < 0.001 576 Cubic Effect • While we are at it, let’s look at the cubic effect – R2 (change) = 0.004, p = 0.045 – 19138 + 103  e + -206  e2 + 12  e3 – Standardised: b1(e) = 0.04 b2(e2) = -2.04 b3(e3) = 2.71 577 Fourth Power • Keep going while we are ahead – won’t run • ??? • Collinearity is the culprit – Tolerance (educ4) = 0.000005 – VIF = 215555 • Matrix of correlations of IVs is not positive definite – cannot be inverted 578 Interpretation • Tricky, given that parameter estimates are a bit nonsensical • Two methods • 1: Use R2 change – Save predicted values • or calculate predicted values to plot line of best fit – Save them from equation – Plot against IV 579 50000 40000 30000 20000 Cubic 10000 Quadratic 0 Linear 8 10 12 14 Education (Years) 16 18 20 22 580 • Differentiate with respect to e • We said: s = 19138 + 103  e + -206  e2 + 12  e3 – but first we will simplify it to quadratic s = 46263 + -6542  e + 310  e2 • dy/dx = -6542 + 310 x 2 x e 581 Education Slope 9 -962 10 -342 11 278 12 898 13 1518 14 2138 15 2758 16 3378 17 3998 18 4618 19 5238 20 5858 1 year of education at the higher end of the scale, better than 1 year at the lower end of the scale. MBA versus GCSE 582 • Differentiate Cubic 19138 + 103  e + -206  e2 + 12  e3 dy/dx = 103 – 206  2  e + 12  3  e2 • Can calculate slopes for quadratic and cubic at different values 583 Education Slope (Quad) Slope (Cub) 9 -962 -689 10 -342 -417 11 278 -73 12 898 343 13 1518 831 14 2138 1391 15 2758 2023 16 3378 2727 17 3998 3503 18 4618 4351 19 5238 5271 20 5858 6263 584 A Quick Note on Differentiation • For y = xp – dx/dy = pxp-1 • For equations such as y =b1x + b2xP dy/dx = b1 + b2pxp-1 • y = 3x + 4x2 – dy/dx = 3 + 4 • 2x 585 • y = b1x + b2x2 + b3x3 – dy/dx = b1 + b2 • 2x + b3 • 3 • x2 • y = 4x + 5x2 + 6x3 • dx/dy = 4 + 5 • 2 • x + 6 • 3 • x2 • Many functions are simple to differentiate – Not all though 586 Automatic Differentiation • If you – Don’t know how to differentiate – Can’t be bothered to look up the function • Can use automatic differentiation software – e.g. GRAD (freeware) 587 588 Lesson 11: Logistic Regression Dichotomous/Nominal Dependent Variables 589 Introduction • Often in social sciences, we have a dichotomous/nominal DV – we will look at dichotomous first, then a quick look at multinomial • Dichotomous DV • e.g. – – – – guilty/not guilty pass/fail won/lost Alive/dead (used in medicine) 590 Why Won’t OLS Do? 591 Example: Passing a Test • Test for bus drivers – pass/fail – we might be interested in degrees of pass fail • a company which trains them will not • fail means ‘pay for them to take it again’ • Develop a selection procedure – Two predictor variables – Score – Score on an aptitude test – Exp – Relevant prior experience (months) 592 • 1st ten cases Score 5 1 1 4 1 1 4 1 3 4 Exp 6 15 12 6 15 6 16 10 12 26 Pass 0 0 0 0 1 0 1 1 0 1 593 • DV – pass (1 = Yes, 0 = No) • Just consider score first – Carry out regression – Score as IV, Pass as DV – R2 = 0.097, F = 4.1, df = 1, 48, p = 0.028. – b0 = 0.190 – b1 = 0.110, p=0.028 • Seems OK 594 • Or does it? … • 1st Problem – pp plot of residuals 1.00 .75 .50 .25 0.00 0.00 .25 Observed Cum Prob .50 .75 1.00 595 • 2nd problem - residual plot 596 • Problems 1 and 2 – strange distributions of residuals – parameter estimates may be wrong – standard errors will certainly be wrong 597 • 3rd problem – interpretation – I score 2 on aptitude. – Pass = 0.190 + 0.110  2 = 0.41 – I score 8 on the test – Pass = 0.190 + 0.110  8 = 1.07 • Seems OK, but – What does it mean? – Cannot score 0.41 or 1.07 • can only score 0 or 1 • Cannot be interpreted – need a different approach 598 A Different Approach Logistic Regression 599 Logit Transformation • In lesson 10, transformed IVs – now transform the DV • Need a transformation which gives us – graduated scores (between 0 and 1) – No upper limit • we can’t predict someone will pass twice – No lower limit • you can’t do worse than fail 600 Step 1: Convert to Probability • First, stop talking about values – talk about probability – for each value of score, calculate probability of pass • Solves the problem of graduated scales 601 probability of failure given a score of 1 is 0.7 Score 1 2 3 4 5 N 7 5 6 4 2 Fail P 0.7 0.5 0.6 0.4 0.2 N 3 5 4 6 8 Pass P 0.3 0.5 0.4 0.6 0.8 probability of passing given a score of 5 is 0.8 602 This is better • Now a score of 0.41 has a meaning – a 0.41 probability of pass • But a score of 1.07 has no meaning – cannot have a probability > 1 (or < 0) – Need another transformation 603 Step 2: Convert to Odds-Ratio Need to remove upper limit • Convert to odds • Odds, as used by betting shops – 5:1, 1:2 • Slightly different from odds in speech – a 1 in 2 chance – odds are 1:1 (evens) – 50% 604 • Odds ratio = (number of times it happened) / (number of times it didn’t happen) p(event) p(event ) odds ratio   p(not event ) 1  p(event ) 605 • 0.8 = 0.8/0.2 = 4 – equivalent to 4:1 (odds on) – 4 times out of five • 0.2 = 0.2/0.8 = 0.25 – equivalent to 1:4 (4:1 against) – 1 time out of five 606 • Now we have solved the upper bound problem – we can interpret 1.07, 2.07, 1000000.07 • But we still have the zero problem – we cannot interpret predicted scores less than zero 607 Step 3: The Log • Log10 of a number(x) log( x ) 10 x • log(10) = 1 • log(100) = 2 • log(1000) = 3 608 • log(1) = 0 • log(0.1) = -1 • log(0.00001) = -5 609 Natural Logs and e • Don’t use log10 – Use loge • Natural log, ln • Has some desirable properties, that log10 doesn’t – – – – For us If y = ln(x) + c dy/dx = 1/x Not true for any other logarithm 610 • Be careful – calculators and stats packages are not consistent when they use log – Sometimes log10, sometimes loge – Can prove embarrassing (a friend told me) 611 Take the natural log of the odds ratio • Goes from -  + – can interpret any predicted value 612 Putting them all together • Logit transformation – log-odds ratio – not bounded at zero or one 613 Score 1 Fail Pass N P N P Odds (Fail) log(odds)fail 2 3 4 7 5 6 4 0.7 0.5 0.6 0.4 3 5 4 6 0.3 0.5 0.4 0.6 5 2 0.2 8 0.8 2.33 1.00 1.50 0.67 0.25 0.85 0.00 0.41 -0.41 -1.39 614 probability 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Probability gets closer to zero, but never reaches it as logit goes down. -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 Logit 615 3.5 • Hooray! Problem solved, lesson over – errrmmm… almost • Because we are now using log-odds ratio, we can’t use OLS – we need a new technique, called Maximum Likelihood (ML) to estimate the parameters 616 Parameter Estimation using ML ML tries to find estimates of model parameters that are most likely to give rise to the pattern of observations in the sample data • All gets a bit complicated – OLS is a special case of ML – the mean is an ML estimator 617 • Don’t have closed form equations – must be solved iteratively – estimates parameters that are most likely to give rise to the patterns observed in the data – by maximising the likelihood function (LF) • We aren’t going to worry about this – except to note that sometimes, the estimates do not converge • ML cannot find a solution 618 Interpreting Output Using SPSS • Overall fit for: – step (only used for stepwise) – block (for hierarchical) – model (always) – in our model, all are the same – c2=4.9, df=1, p=0.025 • F test 619 Om nibus Tests of Model Coe fficients Chi-square St ep 1 df Sig. St ep 4.990 1 .025 Block 4.990 1 .025 Model 4.990 1 .025 620 • Model summary – -2LL (=c2/N) – Cox & Snell R2 – Nagelkerke R2 – Different versions of R2 • No real R2 in logistic regression • should be considered ‘pseudo R2’ 621 Model Sum ma ry St ep 1 -2 Log lik elihood Cox & Snell R Square 64.245 .095 Nagelk erke R Square .127 622 • Classification Table – predictions of model – based on cut-off of 0.5 (by default) – predicted values x actual values 623 Cl assi fication Tablea Predic ted PASS Observed St ep 1 PASS 0 Percentage Correc t 1 0 18 8 69.2 1 12 12 50.0 Overall Percent age 60.0 a. The cut value is .500 624 Model parameters •B – Change in the logged odds associated with a change of 1 unit in IV – just like OLS regression – difficult to interpret • SE (B) – Standard error – Multiply by 1.96 to get 95% CIs 625 Va riables in the Equa tion B Staep 1 S. E. W ald SCORE -.467 .219 4.566 Constant 1.314 .714 3.390 a. Variable(s) ent ered on step 1: SCORE. Variables in the Equation 95.0% C.I.for EXP(B) Sig. Step a 1 Exp(B) s core .386 1.263 Cons tant .199 .323 Lower .744 Upper 2.143 a. Variable(s ) entered on step 1: s core. 626 • Constant – i.e. score = 0 – B = 1.314 – Exp(B) = eB = e1.314 = 3.720 – OR = 3.720, p = 1 – (1 / (OR + 1)) = 1 – (1 / (3.720 + 1)) – p = 0.788 627 • Score 1 – Constant b = 1.314 – Score B = -0.467 – Exp(1.314 – 0.467) = Exp(0.847) = 2.332 – OR = 2.332 – p = 1 – (1 / (2.332 + 1)) = 0.699 628 Standard Errors and CIs • SPSS gives – B, SE B, exp(B) by default – Can work out 95% CI from standard error – B ± 1.96 x SE(B) – Or ask for it in options • Symmetrical in B – Non-symmetrical (sometimes very) in exp(B) 629 Va riables in the Equa tion 95.0% C.I. for EXP(B) B S. E. Ex p(B) SCORE -.467 .219 .627 Constan t 1.314 .714 3.720 Lower .408 Upper .962 a. Variable(s) entered on s tep 1: SCORE. 630 • The odds of passing the test are multiplied by 0.63 (CIs = 0.408, 0.962p p = 0.033), for every additional point on the aptitude test. 631 More on Standard Errors • In OLS regression – If a variable is added in a hierarchical fashion – The p-value associated with the change in R2 is the same as the p-value of the variable – Not the case in logistic regression • In our data 0.025 and 0.033 • Wald standard errors – Make p-value in estimates is wrong – too high – (CIs still correct) 632 • Two estimates use slightly different information – P-value says “what if no effect” – CI says “what if this effect” • Variance depends on the hypothesised ratio of the number of people in the two groups • Can calculate likelihood ratio based pvalues – If you can be bothered – Some packages provide them automatically 633 Probit Regression • Very similar to logistic – much more complex initial transformation (to normal distribution) – Very similar results to logistic (multiplied by 1.7) • In SPSS: – A bit weird • Probit regression available through menus 634 – But requires data structured differently • However – Ordinal logistic regression is equivalent to binary logistic • If outcome is binary – SPSS gives option of probit 635 Results Estimate SE P Logistic (binary) Score 0.288 0.301 0.339 Exp 0.147 0.073 0.043 Logistic (ordinal) Score 0.288 0.301 0.339 Exp 0.147 0.073 0.043 Logistic (probit) Score 0.191 0.178 0.282 Exp 0.090 0.042 0.033 636 Differentiating Between Probit and Logistic • Depends on shape of the error term – Normal or logistic – Graphs are very similar to each other • Could distinguish quality of fit – Given enormous sample size • Logistic = probit x 1.7 – Actually 1.6998 • Probit advantage – Understand the distribution • Logistic advantage – Much simpler to get back to the probability 637 3 2.8 2.6 2.4 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 1 -1 -1.2 -1.4 -1.6 -1.8 -2 -2.2 -2.4 -2.6 -2.8 -3 1.2 Normal (Probit) Logistic 0.8 0.6 0.4 0.2 0 638 Infinite Parameters • Non-convergence can happen because of infinite parameters – Insoluble model • Three kinds: • Complete separation – The groups are completely distinct • Pass group all score more than 10 • Fail group all score less than 10 639 • Quasi-complete separation – Separation with some overlap • Pass group all score 10 or more • Fail group all score 10 or less • Both cases: – No convergence • Close to this – Curious estimates – Curious standard errors 640 • Categorical Predictors – Can cause separation – Esp. if correlated • Need people in every cell Male White Non-White Female White Non-White Below Poverty Line Above Poverty Line 641 Logistic Regression and Diagnosis • Logistic regression can be used for diagnostic tests – For every score • Calculate probability that result is positive • Calculate proportion of people with that score (or lower) who have a positive result • Calculate c statistic – Measure of discriminative power – %age of all possible cases, where the model gives a higher probability to a correct case than to an incorrect case 642 – Perfect c-statistic = 1.0 – Random c-statistic = 0.5 • SPSS doesn’t do it automatically – But easy to do • Save probabilities – Use Graphs, ROC Curve – Test variable: predicted probability – State variable: outcome 643 Sensitivity and Specificity • Sensitivity: – Probability of saying someone has a positive result – • If they do: p(pos)|pos • Specificity – Probability of saying someone has a negative result • If they do: p(neg)|neg 644 Calculating Sens and Spec • For each value – Calculate • proportion of minority earning less – p(m) • proportion of non-minority earning less – p(w) – Sensitivity (value) • P(m) 645 Salary P(minority) 10 20 30 40 50 60 70 80 90 .39 .31 .23 .17 .12 .09 .06 .04 .03 646 Using Bank Data • Predict minority group, using salary (000s) – Logit(minority) = -0.044 + salary x –0.039 • Find actual proportions 647 ROC Curve 1.0 Sensitivity 0.8 0.6 0.4 Area under curve is c-statistic 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 1 - Specificity Diagonal segments are produced by ties. 648 More Advanced Techniques • Multinomial Logistic Regression more than two categories in DV – same procedure – one category chosen as reference group • odds of being in category other than reference • Polytomous Logit Universal Models (PLUM) – Ordinal multinomial logistic regression – For ordinal outcome variables 649 Final Thoughts • Logistic Regression can be extended – dummy variables – non-linear effects – interactions (even though we don’t cover them until the next lesson) • Same issues as OLS – collinearity – outliers 650 651 652 Lesson 12: Mediation and Path Analysis 653 Introduction • Moderator – Level of one variable influences effect of another variable • Mediator – One variable influences another via a third variable • All relationships are really mediated – are we interested in the mediators? – can we make the process more explicit 654 • In examples with bank education beginning salary • Why? – What is the process? – Are we making assumptions about the process? – Should we test those assumptions? 655 job skills expectations beginning salary education negotiating skills kudos for bank 656 Direct and Indirect Influences X may affect Y in two ways • Directly – X has a direct (causal) influence on Y – (or maybe mediated by other variables) • Indirectly – X affects Y via a mediating variable - M 657 • e.g. how does going to the pub effect comprehension on a Summer school course – on, say, regression not reading books on regression Having fun in pub in evening less knowledge Anything here? 658 not reading books on regression Having fun in pub in evening less knowledge fatigue Still needed? 659 • Mediators needed – to cope with more sophisticated theory in social sciences – make explicit assumptions made about processes – examine direct and indirect influences 660 Detecting Mediation 661 4 Steps From Baron and Kenny (1986) • To establish that the effect of X on Y is mediated by M 1. Show that X predicts Y 2. Show that X predicts M 3. Show that M predicts Y, controlling for X 4. If effect of X controlling for M is zero, M is complete mediator of the relationship • (3 and 4 in same analysis) 662 Example: Book habits Enjoy Books  Buy books  Read Books 663 Three Variables • Enjoy – How much an individual enjoys books • Buy – How many books an individual buys (in a year) • Read – How many books an individual reads (in a year) 664 ENJOY BUY READ ENJOY BUY READ 1.00 0.64 0.73 0.64 1.00 0.75 0.73 0.75 1.00 665 • The Theory enjoy buy read 666 • Step 1 1. Show that X (enjoy) predicts Y (read) – b1 = 0.487, p < 0.001 – standardised b1 = 0.732 – OK 667 2. Show that X (enjoy) predicts M (buy) – b1 = 0.974, p < 0.001 – standardised b1 = 0.643 – OK 668 3. Show that M (buy) predicts Y (read), controlling for X (enjoy) – b1 = 0.469, p < 0.001 – standardised b1 = 0.206 – OK 669 4. If effect of X controlling for M is zero, M is complete mediator of the relationship – (Same as analysis for step 3.) – b2 = 0.287, p = 0.001 – standardised b2 = 0.431 – Hmmmm… • Significant, therefore not a complete mediator 670 0.287 (step 4) enjoy read buy 0.974 (from step 2) 0.206 (from step 3) 671 The Mediation Coefficient • Amount of mediation = Step 1 – Step 4 =0.487 – 0.287 = 0.200 • OR Step 2 x Step 3 =0.974 x 0.206 = 0.200 672 SE of Mediator enjoy buy a (from step 2) read b (from step 2) • sa = se(a) • sb = se(b) 673 • Sobel test – Standard error of mediation coefficient can be calculated se  b s + a s - s s 2 2 a a = 0.974 sa = 0.189 2 2 b 2 2 a b b = 0.206 sb = 0.054 674 • Indirect effect = 0.200 – se = 0.056 – t =3.52, p = 0.001 • Online Sobel test: http://www.unc.edu/~preacher/sobel/ sobel.htm – (Won’t be there for long; probably will be somewhere else) 675 A Note on Power • Recently – Move in methodological literature away from this conventional approach – Problems of power: – Several tests, all of which must be significant • Type I error rate = 0.05 * 0.05 = 0.0025 • Must affect power – Bootstrapping suggested as alternative • See Paper B7, A4, B9 • B21 for SPSS syntax 676 677 678 Lesson 13: Moderators in Regression “different slopes for different folks” 679 Introduction • Moderator relationships have many different names – interactions (from ANOVA) – multiplicative – non-linear (just confusing) – non-additive • All talking about the same thing 680 A moderated relationship occurs • when the effect of one variable depends upon the level of another variable 681 • Hang on … – That seems very like a nonlinear relationship – Moderator • Effect of one variable depends on level of another – Non-linear • Effect of one variable depends on level of itself • Where there is collinearity – Can be hard to distinguish between them – Paper in handbook (B5) – Should (usually) compare effect sizes 682 • e.g. How much it hurts when I drop a computer on my foot depends on – x1: how much alcohol I have drunk – x2: how high the computer was dropped from – but if x1 is high enough – x2 will have no effect 683 • e.g. Likelihood of injury in a car accident – depends on – x1: speed of car – x2: if I was wearing a seatbelt – but if x1 is low enough – x2 will have no effect 684 30 25 Injury 20 15 10 5 0 5 15 25 35 45 Speed (mph) Seatbelt No Seatbelt 685 • e.g. number of words (from a list) I can remember – depends on – x1: type of words (abstract, e.g. ‘justice’, or concrete, e.g. ‘carrot’) – x2: Method of testing (recognition – i.e. multiple choice, or free recall) – but if using recognition – x1: will not make a difference 686 • We looked at three kinds of moderator • alcohol x height = pain – continuous x continuous • speed x seatbelt = injury – continuous x categorical • word type x test type – categorical x categorical • We will look at them in reverse order 687 How do we know to look for moderators? Theoretical rationale • Often the most powerful • Many theories predict additive/linear effects – Fewer predict moderator effects Presence of heteroscedasticity • Clue there may be a moderated relationship missing 688 Two Categorical Predictors 689 • 2 IVs Data – word type (concrete [1], abstract [2]) – test method (recog [1], recall [2]) • 20 Participants in one of four groups – – – – 1, 1, 2, 2, 1 2 1 2 • 5 per group • lesson12.1.sav 690 Recog Recall Total Concrete Abstract Total Mean 15.40 15.20 15.30 SD 2.19 2.59 2.26 Mean 15.60 6.60 11.10 Std. Deviation 1.67 7.44 6.95 Mean 15.50 10.90 13.20 Std. Deviation 1.84 6.94 5.47 691 • Graph of means 18 16 14 12 10 W ORDS 8 1.00 6 1.00 2.00 2.00 TEST 692 ANOVA Results • Standard way to analyse these data would be to use ANOVA – Words: F=6.1, df=1, 16, p=0.025 – Test: F=5.1, df=1, 16, p=0.039 – Words x Test: F=5.6, df=1, 16, p=0.031 693 Procedure for Testing 1: Convert to effect coding • can use dummy coding, collinearity is less of an issue • doesn’t make any difference to substantive interpretation 2: Calculate interaction term • In ANOVA interaction is automatic • In regression we create an interaction variable 694 • Interaction term (wxt) – multiply effect coded variables together word -1 1 -1 1 test -1 -1 1 1 wxt 1 -1 -1 1 695 3: Carry out regression • Hierarchical – linear effects first – interaction effect in next block 696 b0=13.2 b1 (words) = -2.3, p=0.025 b2 (test) = -2.1, p=0.039 b3 (words x test) = -2.2, p=0.031 Might need to use change in R2 to test sig of interaction, because of collinearity What do these mean? • b0 (intercept) = predicted value of Y (score) when all X = 0 • • • • • – i.e. the central point 697 • b0 = 13.2 – grand mean • b1 = -2.3 – distance from grand to mean for two word types – 13.2 – (-2.3) = 15.5 – 13.2 + (-2.3) = 10.9 Recog Recall Total Concrete Abstract Total 15.40 15.20 15.30 15.60 6.60 11.10 15.50 10.90 13.20 698 • b2 = -2.1 – distance from grand mean to recog and recall means • b3 = -2.2 – to understand b3 we need to look at predictions from the equation without this term Score = 13.2 + (-2.3)  w + (-2.1)  t 699 Score = 13.2 + (-2.3)  w + (-2.1)  t • So for each group we can calculate an expected value 700 b1 = -2.3, b2 = -2.1 W T Word Test Expected Value C Cog -1 -1 13.2 + (-2.3) x (-1) + (-2.1) x -1 C Call -1 1 13.2 + (-2.3) x (-1) + (-2.1) x 1 A Cog 1 -1 13.2 + (-2.3) x 1 + (-2.1) x (-1) A Call 1 1 13.2 + (-2.3) x 1 + (-2.1) x 1 701 W C C A A T Word Test Exp Actual Value Call -1 -1 17.6 15.4 Cog -1 1 13.4 15.6 Call 1 -1 13.0 15.2 Cog 1 1 8.8 11.0 • The exciting part comes when we look at the differences between the actual value and the value in the 2 IV model 702 • Each difference = 2.2 (or –2.2) • The value of b3 was –2.2 – the interaction term is the correction required to the slope when the second IV is included 703 • Examine the slope for word type 18 16 14 12 10 8 6 4 Gradient = (11.1 - 15.3) / 2 = 2.1 2 0 Recog (-1) Recall (1) Test Type 704 • Add the slopes for two test groups 18 16 14 12 10 8 Both word groups (-2.1) 6 4 2 0 Recog (-1) Abstract (6.6 - 15.2 )/2 = -4.3 Concrete (15.6-15.4 )/2 = 0.1 Recall (1) Test Type 705 b associated with interaction • the change in slope, away from the average, associated with a 1 unit change in the moderating variable OR • Half the difference in the slopes 706 • Another way to look at it Y = 13.2 + -2.3w + -2.1t + -2.2wt • Examine concrete words group (w = -1) – substitute values into the equation Y(concrete) = 13.2 + -2.3-1 + -2.1t + -2.2-1t Y(concrete) = 13.2 + 2.3 + -2.1t + 2.2t Y(concrete) = 15.5 + 0.1t • The effect of changing test type for concrete words (the slope, which is half the actual difference) 707 Why go to all that effort? Why not do ANOVA in the first place? 1. That is what ANOVA actually does • • • if it can handle an unbalanced design (i.e. different numbers of people in each group) Helps to understand what can be done with ANOVA SPSS uses regression to do ANOVA 2. Helps to clarify more complex cases • as we shall see 708 Categorical x Continuous 709 Note on Dichotomisation • Very common to see people dichotomise a variable – Makes the analysis easier – Very bad idea • Paper B6 710 Data A chain of 60 supermarkets • examining the relationship between profitability, shop size, and local competition • 2 IVs – shop size – comp (local competition, 0=no, 1=yes) • DV – profit 711 • Data, ‘lesson 12.2.sav’ Shopsize 4 10 7 10 10 29 12 6 14 62 Comp 1 1 0 0 1 1 0 1 0 0 Profit 23 25 19 9 18 33 17 20 21 8 712 1st Analysis Two IVs • R2=0.367, df=2, 57, p < 0.001 • Unstandardised estimates – b1 (shopsize) = 0.083 (p=0.001) – b2 (comp) = 5.883 (p<0.001) • Standardised estimates – b1 (shopsize) = 0.356 – b2 (comp) = 0.448 713 • Suspicions – Presence of competition is likely to have an effect – Residual plot shows a little heteroscedasticity 3 2 1 0 -1 -2 -3 -2.0 -1.5 -1.0 -.5 0.0 .5 1.0 1.5 2.0 714 Procedure for Testing • Very similar to last time – convert ‘comp’ to effect coding – -1 = No competition – 1 = competition – Compute interaction term • comp (effect coded) x size – Hierarchical regression 715 Result • Unstandardised estimates – b1 (shopsize) = 0.071 (p=0.006) – b2 (comp) = -1.67 (p = 0.506) – b3 (sxc) = -0.050 (p=0.050) • Standardised estimates – b1 (shopsize) = 0.306 – b2 (comp) = -0.127 – b3 (sxc) = -0.389 716 • comp now non-significant – shows importance of hierarchical – it obviously is important 717 Interpretation • Draw graph with lines of best fit – drawn automatically by SPSS • Interpret equation by substitution of values – evaluate effects of • size • competition 718 40 30 20 10 Profit Competition No competition 0 All Shops 0 20 40 60 80 100 Shopsize 719 • Effects of size – in presence and absence of competition – (can ignore the constant) Y=x10.071 + x2(-1.67) + x1x2 (-0.050) – Competition present (x2 = 1) Y=x10.071 + 1(-1.67) + x11 (-0.050) Y=x10.071 + -1.67 + x1(-0.050) Y=x1 0.021 + (–1.67) 720 Y=x10.071 + x2(-1.67) + x1x2 (-0.050) – Competition absent (x2 = -1) Y=x10.071 + -1(-1.67) + x1-1 (-0.050) Y=x1 0.071 + x1-1 (-0.050) + -1(-1.67) Y= x1 0.121 (+ 1.67) 721 Two Continuous Variables 722 Data • Bank Employees – only using clerical staff – 363 cases – predicting starting salary – previous experience – age – age x experience 723 • Correlation matrix – only one significant LOGSB AGESTARTPREVEXP LOGSB 1.00 -0.09 0.08 AGESTART -0.09 1.00 0.77 PREVEXP 0.08 0.77 1.00 724 Initial Estimates (no moderator) • (standardised) – R2 = 0.061, p<0.001 – Age at start = -0.37, p<0.001 – Previous experience = 0.36, p<0.001 • Suppressing each other – Age and experience compensate for one another – Older, with no experience, bad – Younger, with experience, good 725 The Procedure • Very similar to previous – create multiplicative interaction term – BUT • Need to eliminate effects of means – cause massive collinearity • and SDs – cause one variable to dominate the interaction term • By standardising 726 • To standardise x, – subtract mean, and divide by SD – re-expresses x in terms of distance from the mean, in SDs – ie z-scores • Hint: automatic in SPSS in Descriptives • Create interaction term of age and exp – axe = z(age)  z(exp) 727 • Hierarchical regression – two linear effects first – moderator effect in second – hint: it is often easier to interpret if standardised versions of all variables are used 728 • Change in R2 – 0.085, p<0.001 • Estimates (standardised) – b1 (exp) = 0.104 – b2 (agestart) = -0.54 – b3 (age x exp) = -0.54 729 Interpretation 1: Pick-a-Point • Graph is tricky – can’t have two continuous variables – Choose specific points (pick-a-point) • Graph the line of best fit of one variable at others – Two ways to pick a point • 1: Choose high (z = +1), medium (z = 0) and low (z = -1) • Choose ‘sensible’ values – age 20, 50, 80? 730 • We know: – Y = e  0.10 + a  -0.54 + a  e  -0.54 – Where a = agestart, and e = experience • We can rewrite this as: – Y = (e  0.10) + (a  -0.54) + (a  e  -0.54) – Take a out of the brackets – Y = (e  0.10) + (-0.54 + e  -0.54)a • Bracketed terms are simple intercept and simple slope – 0= (e  0.10) – 1= (-0.54 + e  -0.54)a – Y = 0 + 1a 731 • Pick any value of e, and we know the slope for a – Standardised, so it’s easy • e = -1 – 0= (-1  0.10) = -0.10 – 1= (-0.54 + -1  -0.54)a = -0.0a • e=0 – 0= (0  0.10) = 0 – 1= (-0.54+ 0  -0.54)a = -0.54a • e=1 – 0= (1  0.10) = 0.10 – 1= (-0.54 + 1  -0.54)a = -1.08a 732 Graph the Three Lines 1.5 1 e = -1 e=0 e=1 Log(salary) 0.5 0 -0.5 -1 -1.5 -1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 Age 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 733 Interpretation 2: P-Values and CIs • Second way – Newer, rarely done • Calculate CIs of the slope – At any point • Calculate p-value – At any point • Give ranges of significance 734 What do you need? • The variance and covariance of the estimates – SPSS doesn’t provide estimates for intercept – Need to do it manually • In options, exclude intercept – Create intercept – c = 1 – Use it in the regression 735 • Enter information into web page: – www.unc.edu/~preacher/interact/a cov.htm – (Again, may not be around for long) • Get results • Calculations in Bauer and Curran (in press: Multivariate Behavioral Research) – Paper B13 736 4.1 4.2 Y 4.3 4.4 4.5 MLR 2-Way Interaction Plot 4.0 CVz1(1) CVz1(2) CVz1(3) -1.0 -0.5 0.0 X 0.5 1.0 737 Areas of Significance 0.0 -0.2 -0.4 -0.6 Simple Slope 0.2 0.4 Confidence Bands -4 -2 0 2 4 Experience 738 • 2 complications – 1: Constant differed – 2: DV was logged, hence non-linear • effect of 1 unit depends on where the unit is – Can use SPSS to do graphs showing lines of best fit for different groups – See paper A2 739 Finally … 740 Unlimited Moderators • Moderator effects are not limited to – 2 variables – linear effects 741 Three Interacting Variables • Age, Sex, Exp • Block 1 – Age, Sex, Exp • Block 2 – Age x Sex, Age x Exp, Sex x Exp • Block 3 – Age x Sex x Exp 742 • Results – All two way interactions significant – Three way not significant – Effect of Age depends on sex – Effect of experience depends on sex – Size of the age x experience interaction does not depend on sex (phew!) 743 Moderated Non-Linear Relationships • Enter non-linear effect • Enter non-linear effect x moderator – if significant indicates degree of nonlinearity differs by moderator 744 745 Modelling Counts: Poisson Regression Lesson 14 746 Counts and the Poisson Distribution • Von Bortkiewicz (1898) – Numbers of Prussian soldiers kicked to death by horses 120 100 80 60 0 1 2 3 4 5 109 65 22 3 1 0 40 20 0 0 1 2 3 4 5 747 • The data fitted a Poisson probability distribution – When counts of events occur, poisson distribution is common – E.g. papers published by researchers, police arrests, number of murders, ship accidents • Common approach – Log transform and treat as normal • Problems – Censored at 0 – Integers only allowed – Heteroscedasticity 748 The Poisson Distribution 0.7 0.6 Probability 0.5 0.5 1 4 8 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 Count 10 11 12 13 14 15 16 17 749 exp(   )  p ( y | x)  y! y 750 exp(  )  p ( y | x)  y! y • Where: – y is the count –  is the mean of the poisson distribution • In a poisson distribution – The mean = the variance (hence heteroscedasticity issue)) –   2 751 Poisson Regression in SPSS • Not directly available – SPSS can be tweaked to do it in three ways: – General loglinear model (genlog) – Non-linear regression (CNLR) • Bootstrapped p-values only – Both are quite tricky • SPSS 15, 752 Example Using Genlog – 100 surfboards, 50 red, 50 blue • Weight cases by bites • Analyse, Loglinear, General – Colour is factor 25 20 Blue Red 15 Frequency • Number of shark bites on different colour surfboards 10 5 0 0 1 2 Number of bites 3 4 753 Results Correspondence Between Parameters and Terms of the Design Parameter Aliased Term 1 Constant 2 [COLOUR = 1] 3 x [COLOUR = 2] Note: 'x' indicates an aliased (or a redundant) parameter. These parameters are set to zero. 754 Asymptotic Param Est. 1 2 3 4.1190 -.5495 .0000 SE .1275 .2108 . Z-value 95% CI Lower Upper 32.30 -2.61 . 3.87 -.96 . 4.37 -.14 . • Note: Intercept (param 1) is curious • Param 2 is the difference in the means 755 SPSS: Continuous Predictors • Bleedin’ nightmare • http://www.spss.com/tech/answer/detai ls.cfm?tech_tan_id=100006204 756 Poisson Regression in Stata • SPSS will save a Stata file • Open it in Stata • Statistics, Count outcomes, Poisson regression 757 Poisson Regression in R • R is a freeware program – Similar to SPlus – www.r-project.org • Steep learning curve to start with • Much nicer to do Poisson (and other) regression analysis http://www.stat.lsa.umich.edu/~faraway/book / http://www.jeremymiles.co.uk/regressionbook /extras/appendix2/R/ 758 • Commands in R • Stage 1: enter data – colour <- c(1, 0, 1, 0, 1, 0 … 1) – bites <- c(3, 1, 0, 0, … ) • Run analysis – p1 <- glm(bites ~ colour, family = poisson) • Get results – summary.glm(p1) 759 R Results Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -0.3567 0.1686 -2.115 0.03441 * colour 0.5555 0.2116 2.625 0.00866 ** • Results for colour – Same as SPSS – For intercept different (weird SPSS) 760 Predicted Values • Need to get exponential of parameter estimates – Like logistic regression • Exp(0.555) = 1.74 – You are likely to be bitten by a shark 1.74 times more often with a red surfboard 761 Checking Assumptions • Was it really poisson distributed? – For Poisson,   2 • As mean increases, variance should also increase – Residuals should be random • Overdispersion is common problem • Too many zeroes • For blue:   2 = exp(-0.3567) = 1.42 • For red:   2 = exp(-0.3567 + 0.555) = 2.48 762 exp(  )  p ( y | x)  y! y • Strictly: exp( ˆ ) ˆ p( yi | xi )  y! y 763 Compare Predicted with Actual Distributions Red Blue 0.7 0.4 0.6 0.35 0.3 Probability Expected Actual 0.4 0.3 0.25 Probability 0.5 0.2 0.15 0.2 Expected Actual 0.1 0.1 0.05 0 0 0 1 2 Frequency 3 4 1 2 3 4 Frequency 764 Overdispersion • Problem in poisson regression – Too many zeroes • Causes – c2 inflation – Standard error deflation • Hence p-values too low – Higher type I error rate • Solution – Negative binomial regression 765 Using R • R can read an SPSS file – But you have to ask it nicely • Click Packages menu, Load package, choose “Foreign” • Click File, Change Dir – Change to the folder that contains your data 766 More on R • R uses objects – To place something into an object use <– X <- Y • Puts Y into X • Function is read.spss() – Mydata <- read.spss(“spssfilename.sav”) • Variables are then referred to as Mydata$VAR1 – Note 1: R is case sensitive – Note 2: SPSS variable name in capitals 767 GLM in R • Command – glm(outcome ~ pred1 + pred2 + … + predk [,family = familyname]) – If no familyname, default is OLS • Use binomial for logistic, poisson for poisson • Output is a GLM object – You need to give this a name – my1stglm <- glm(outcome ~ pred1 + pred2 + … + predk [,family = familyname]) 768 • Then need to explore the result – summary(my1stglm) • To explore what it means – Need to plot regressions • Easiest is to use Excel 769 770 Introducing Structural Equation Modelling Lesson 15 771 Introduction • Related to regression analysis – All (OLS) regression can be considered as a special case of SEM • Power comes from adding restrictions to the model • SEM is a system of equations – Estimate those equations 772 Regression as SEM • Grades example – Grade = constant + books + attend + error • Looks like a regression equation – Also – Books correlated with attend – Explicit modelling of error 773 Path Diagram • System of equations are usefully represented in a path diagram x Measured variable e unmeasured variable regression correlation 774 Path Diagram for Regression Must usually explicitly model error error Books Grade Attend Must explicitly model correlation 775 Results • Unstandardised 2.00 1.00 e BOOKS 4.04 2.65 13.52 GRADE 17.84 1.28 ATTEND 776 Standardised e BOOKS .35 .44 .82 GRADE .33 ATTEND 777 Table GRADE GRADE GRADE GRADE <-- BOOKS <-- ATTEND <-- e Estimate 4.04 1.28 13.52 37.38 S.E. 1.71 0.57 1.53 7.54 C.R. 2.36 2.25 8.83 4.96 P St. Est. 0.02 0.35 0.03 0.33 0.00 0.82 0.00 Coeffi ci entsa Unstandardized Coefficient s M odel 1 B Standardized Coefficient s Std. Error Beta Sig. 37.38 7.74 BOOKS 4.04 1.75 .35 .03 ATTEND 1.28 .59 .33 .04 (Const ant) a. Dep endent Variable: GRADE .00 778 So What Was the Point? • Regression is a special case • Lots of other cases • Power of SEM – Power to add restrictions to the model • Restrict parameters – To zero – To the value of other parameters – To 1 779 Restrictions • Questions – Is a parameter really necessary? – Are a set of parameters necessary? – Are parameters equal • Each restriction adds 1 df – Test of model with c2 780 The c2 Test • Can the model proposed have generated the data? – Test of significance of difference of model and data – Statistically significant result • Bad – Theoretically driven • Start with model • Don’t start with data 781 Regression Again 0, 1 BOOKS e GRADE ATTEND • Both estimates restricted to zero 782 • Two restrictions – 2 df for c2 test – c2 = 15.9, p = 0.0003 • This test is (asymptotically) equivalent to the F test in regression – We still haven’t got any further 783 Multivariate Regression y1 x1 y2 x2 y3 784 Test of all x’s on all y’s (6 restrictions = 6 df) y1 x1 y2 x2 y3 785 Test of all x1 on all y’s (3 restrictions) y1 x1 y2 x2 y3 786 Test of all x1 on all y1 (3 restrictions) y1 x1 y2 x2 y3 787 Test of all 3 partial correlations between y’s, controlling for x’s (3 restrictions) y1 x1 y2 x2 y3 788 Path Analysis and SEM • More complex models – can add more restrictions ENJOY 1 – E.g. mediator model BUY e_buy • 1 restriction – No path from enjoy -> read 1 READ e_read 789 Result • c2 = 10.9, 1 df, p = 0.001 • Not a complete mediator – Additional path is required 790 Multiple Groups • Same model – Different people • Equality constraints between groups – Means, correlations, variances, regression estimates – E.g. males and females 791 Multiple Groups Example • Age • Severity of psoriasis – SEVE – in emotional areas • Hands, face, forearm – SEVNONE – in non-emotional areas – Anxiety – Depression 792 Correlationsa AGE AGE .017 .035 . .004 .009 .859 .717 110 110 110 110 110 -.270 1 .665 .045 .075 .004 . .000 .639 .436 110 110 110 110 110 -.248 .665 1 .109 .096 .009 .000 . .255 .316 110 110 110 110 110 Pearson Correlat ion .017 .045 .109 1 .782 Sig. (2-tailed) .859 .639 .255 . .000 110 110 110 110 110 Pearson Correlat ion .035 .075 .096 .782 1 Sig. (2-tailed) .717 .436 .316 .000 . 110 110 110 110 110 Pearson Correlat ion N Pearson Correlat ion Sig. (2-tailed) N N GHQ_D GHQ_D -.248 Sig. (2-tailed) GHQ_A GHQ_A -.270 N SEVNONE SEVNONE 1 Pearson Correlat ion Sig. (2-tailed) SEVE SEVE N a. SEX = f 793 Correlationsa AGE AGE Pearson Correlat ion Sig. (2-tailed) N SEVE Pearson Correlat ion Sig. (2-tailed) N SEVNONE Pearson Correlat ion Sig. (2-tailed) N GHQ_A Pearson Correlat ion Sig. (2-tailed) N GHQ_D Pearson Correlat ion Sig. (2-tailed) N SEVE SEVNONE GHQ_A GHQ_D 1 -.243 -.116 -.195 -.190 . .031 .310 .085 .094 79 79 79 79 79 -.243 1 .671 .456 .453 .031 . .000 .000 .000 79 79 79 79 79 -.116 .671 1 .210 .232 .310 .000 . .063 .040 79 79 79 79 79 -.195 .456 .210 1 .800 .085 .000 .063 . .000 79 79 79 79 79 -.190 .453 .232 .800 1 .094 .000 .040 .000 . 79 79 79 79 79 a. SEX = m 794 Model AGE SEVE SEVNONE 1 1 e_s e_sn Dep Anx 1 1 E_d e_a 795 Females AGE -.27 -.25 SEVE .96 SEVNONE .07 .04 e_s .97 e_sn .03 .09 -.04 .15 .64 Dep Anx .99 .99 E_d e_a .78 796 AGE Males -.24 -.12 SEVE .97 SEVNONE -.08 -.08 e_s .99 e_sn .52 -.12 .55 -.17 .67 Dep Anx .88 .88 E_d e_a .74 797 Constraint • sevnone -> dep – Constrained to be equal for males and females • 1 restriction, 1 df – c2 = 1.3 – not significant • 4 restrictions – 2 severity -> anx & dep 798 • 4 restrictions, 4 df – c2 = 1.3, p = 0.014 • Parameters are not equal 799 Missing Data: The big advantage • SEM programs tend to deal with missing data – Multiple imputation – Full Information (Direct) Maximum Likelihood • Asymptotically equivalent • Data can be MAR, not just MCAR 800 Power: A Smaller Advantage • Power for regression gets tricky with large models • With SEM power is (relatively) easy – It’s all based on chi-square – Paper B14 801 Lesson 16: Dealing with clustered data & longitudinal models 802 The Independence Assumption • In Lesson 8 we talked about independence – The residual of any one case should not tell you about the residual of any other case • Particularly problematic when: – Data are clustered on the predictor variable • E.g. predictor is household size, cases are members of family • E.g. Predictor is doctor training, outcome is patients of doctor – Data are longitudinal • Have people measured over time – It’s the same person! 803 Clusters of Cases • Problem with cluster (group) randomised studies – Or group effects • Use Huber-White sandwich estimator – Tell it about the groups – Correction is made – Use complex samples in SPSS 804 Complex Samples • As with Huber-White for heteroscedasticity – Add a variable that tells it about the clusters – Put it into clusters • Run GLM – As before • Warning: – Need about 20 clusters for solutions to be stable 805 Example • People randomised by week to one of two forms of triage – Compare the total cost of treating each • Ignore clustering – Difference is £2.40 per person, with 95% confidence intervals £0.58 to £4.22, p =0.010 • Include clustering – Difference is still £2.40, with 95% CIs £5.65 to £0.85, and p = 0.141. • Ignoring clustering led to type I error 806 Longitudinal Research • For comparing repeated measures – Clusters are people – Can model the repeated measures over time ID V1 V2 V3 V4 1 2 3 4 7 2 3 6 8 4 3 2 5 7 5 • Data are usually short and fat 807 Converting Data • Change data to tall and thin • Use Data, Restructure in SPSS • Clusters are ID ID V X 1 1 2 1 2 3 1 3 4 1 4 7 2 1 3 2 2 6 2 3 8 2 4 4 3 1 2 3 2 5 3 3 7 3 4 5 808 (Simple) Example • Use employee data.sav – Compare beginning salary and salary – Would normally use paired samples t-test • Difference = $17,403, 95% CIs $16,427.407, $18,379.555 809 Restructure the Data • Do it again – With data tall and thin • Complex GLM with Time as factor – ID as cluster • Difference = $17,430, 95% CIs = 16427.407, 18739.555 ID Time Cash 1 1 $18,750 1 2 $21,450 2 1 $12,000 2 2 $21,900 3 1 $13,200 3 2 $45,000 810 Interesting … • That wasn’t very interesting – What is more interesting is when we have multiple measurements of the same people • Can plot and assess trajectories over time 811 Single Person Trajectory + + + + + + Time 812 Multiple Trajectories: What’s the Mean and SD? Time 813 Complex Trajectories • An event occurs – Can have two effects: – A jump in the value – A change in the slope • Event doesn’t have to happen at the same time for each person – Doesn’t have to happen at all 814 Slope 1 Jump Slope 2 Event Occurs 815 Parameterising Time 1 2 3 4 5 6 7 8 9 Event 0 0 0 0 0 1 1 1 1 Time2 0 0 0 0 0 0 1 2 3 Outcome 12 13 14 15 16 10 9 8 7 816 Draw the Line What are the parameter estimates? 817 Main Effects and Interactions • Main effects – Intercept differences • Moderator effects – Slope differences 818 Multilevel Models • Fixed versus random effects – Fixed effects are fixed across individuals (or clusters) – Random effects have variance • Levels – Level 1 – individual measurement occasions – Level 2 – higher order clusters 819 More on Levels • NHS direct study – Level 1 units: ……………. – Level 2 units: …………… • Widowhood food study – Level 1 units …………… – Level 2 units …………… 820 More Flexibility • Three levels: – Level 1: measurements – Level 2: people – Level 3: schools 821 More Effects • Variances and covariances of effects • Level 1 and level 2 residuals – Makes R2 difficult to talk about • Outcome variable – Yij • The score of the ith person in the jth group 822 Y 2.3 3.2 4.5 4.8 7.2 3.1 1.6 i 1 2 3 1 2 3 4 j 1 1 1 2 2 2 2 823 Notation • Notation gets a bit horrid – Varies a lot between books and programs • We used to have b0 and b1 – If fixed, that’s fine – If random, each person has their own intercept and slope 824 Standard Errors • Intercept has standard errors • Slopes have standard errors • Random effects have variances – Those variances have standard errors • Is there statistically significant variation between higher level units (people)? • OR • Is everyone the same? 825 Programs • Since version 12 – Can do this in SPSS – Can’t do anything really clever • Menus – Completely unusable – Have to use syntax 826 SPSS Syntax • MIXED • relfd with time • /fixed = time • /random = intercept time | subject (id) covtype(un) • /print = solution. 827 SPSS Syntax • MIXED • relfd with time Outcome Continuous predictor 828 SPSS Syntax • MIXED • relfd with time • /fixed = time Must specify effect as fixed first 829 SPSS Syntax • MIXED • relfd with time • /fixed = time • /random = intercept time | subject Intercept and (id) covtype(un) time are random Specify random effects SPSS assumes that your level 2 units are subjects, and needs to know the id variable 830 SPSS Syntax • MIXED • relfd with time • fixed = time • /random = intercept time | subject (id) covtype(un) Covariance matrix of random effects is unstructured. (Alternative is id – identity or vc – variance components). 831 SPSS Syntax • MIXED • relfd with time • fixed = time • /random = intercept time | subject (id) covtype(un) • /print = solution. Print the answer 832 The Output • Information criteria – We’ll come back Information Criteriaa -2 Restricted Log Likelihood 64899.758 Akaike's Information 64907.758 Criterion (AIC) Hurvich and Ts ai's Criterion (AICC) 64907.763 Bozdogan's Criterion 64940.134 (CAIC) Schwarz's Bayes ian 64936.134 Criterion (BIC) The information criteria are displayed in smaller-is -better forms . a. Dependent Variable: relfd. 833 Fixed Effects • Not useful here, useful for interactions Type III Tests of Fixed Effectsa Numerator df Denominator df Intercept 1 741 3251.877 .000 time 1 741.000 2.550 .111 Source F Sig. a. Dependent Variable: relfd. 834 Estimates of Fixed Effects • Interpreted as regression equation Estimates of Fixed Effectsa 95% Confidence Interval Parameter Intercept time Estimate Std. Error df t Sig. Lower Bound Upper Bound 21.90 21.90 .38 57.025 .000 21.15 22.66 -.06 -.06 .04 -1.597 .111 -.14 .01 a. Dependent Variable: relfd. 835 Covariance Parameters Estimates of Covariance Parametersa Parameter Estimate Res idual 64.11577 1.0526353 Intercept + time [subject = id] Std. Error UN (1,1) 85.16791 5.7003732 UN (2,1) -4.53179 .5067146 UN (2,2) .7678319 .0636116 a. Dependent Variable: relfd. 836 Change Covtype to VC • We know that this is wrong – The covariance of the effects was statistically significant – Can also see if it was wrong by comparing information criteria • We have removed a parameter from the model – Model is worse – Model is more parsimonious • Is it much worse, given the increase in parsimony? 837 UN Model Information Criteriaa -2 Restricted Log Likelihood 64899.758 VC Model Information Criteriaa -2 Restricted Log Likelihood 65041.891 Akaike's Information 64907.758 Criterion (AIC) Akaike's Information 65047.891 Criterion (AIC) Hurvich and Ts ai's Criterion (AICC) Hurvich and Ts ai's Criterion (AICC) 64907.763 65047.894 Bozdogan's Criterion 64940.134 (CAIC) Bozdogan's Criterion 65072.173 (CAIC) Schwarz's Bayes ian 64936.134 Criterion (BIC) Schwarz's Bayes ian 65069.173 Criterion (BIC) The information criteria are displayed in smaller-is The information criteria are displayed in smaller-is -better forms . a. Dependent Variable: relfd. a. Dependent Variable: relfd. Lower is better. 838 Adding Bits • So far, all a bit dull • We want some more predictors, to make it more exciting – E.g. female – Add: Relfd with time female /fixed = time sex time * sex • What does the interaction term represent? 839 Extending Models • Models can be extended – Any kind of regression can be used • Logistic, multinomial, Poisson, etc – More levels • Children within classes within schools • Measures within people within classes within prisons – Multiple membership / cross classified models • Children within households and classes, but households not nested within class • Need a different program – E.g. MlwiN 840 MlwiN Example (very quickly) 841 Books Singer, JD and Willett, JB (2003). Applied Longitudinal Data Analysis: Modeling Change and Event Occurrence. Oxford, Oxford University Press. Examples at: http://www.ats.ucla.edu/stat/SPSS/ex amples/alda/default.htm 842 The End 843

Top subcategories

Top subcategories

Top subcategories

Top subcategories

Top subcategories

Top subcategories

Top subcategories

Top subcategories

Download Theory of Regression