Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Section 7.1 1 An unreasonable assumption A major drawback of our inference methods so far is that s is almost never known in practice. We can estimate it from the sample by replacing it with the estimate s, but this estimate is often poor, especially when n is small. • standard error the statistic s n , used as an estimate for the standard deviation s n of the sampling distribution for the sample mean statistic x † † † Section 7.1 2 The t-distribution When we replace s with the estimate s in the formula for the statistic z= x -m 0, s n we obtain a new statistic † x -m 0 t= s n which has a distribution that is not normal, although it † it is bell-shaped and symmetric about 0 like the z distribution N(0,!1). However, the t distribution • has thicker tails (larger spread) than the standard normal distribution (substituting s with s introduces more variation) • becomes closer to a normal distribution as the number of degrees of freedom (n!–!1) increases (increasing n leads to a closer estimate of s by s) Section 7.1 3 One-sample t procedures From a SRS selected from a normal population, or a large SRS from any population, determine sample statistics x and s. A level C confidence interval for m is † x ± t* s n 1-C critical value for the 2 t-distribution with n – 1 degrees of freedom † where t* is the upper † † Section 7.1 4 One-sample t hypothesis test Assumptions: SRS selected from a normal population, or a large SRS from any population • State hypotheses: Null hypothesis Alternative hypothesis H0: m = m0 Ha: m > m0, or m < m0, or m ≠ m0 • Calculate test statistic: t-statistic based on H0: x -m 0 t= s n • Find P-value: Sampling distribution probability associated with † appropriate H0: P = P( T ≥ t ), or P = P( T ≤ t ), or P = 2P( T ≥ t ) Conclusion: assess evidence against H0 in favor of Ha depending on how small P is. [TI-83: STAT TESTS T-Test… ] Section 7.1 5 Matched pairs procedures To compare responses to two treatments in a matched pairs design, apply the one-sample t procedures to the observed differences of the pairs. • robustness statistical inference technique in which the level of confidence or P-value change little when the underlying assumptions are violated Since the t-distributions have thick tails, outliers are more common than for normal distributions, so like x, the t procedures are strongly influenced by outliers. † Consequently, t procedures are robust when • no outliers are present, • the data are symmetrically distributed, or • when sample sizes are large.