Chapter 15 Comparing Two Populations Dependent samples Dependent

  • Slides: 49
Download presentation
Chapter 15 Comparing Two Populations: Dependent samples

Chapter 15 Comparing Two Populations: Dependent samples

Dependent Samples l Dependent Samples - measurements are somehow related to one another either

Dependent Samples l Dependent Samples - measurements are somehow related to one another either because: – The measurements came from the same subject (Within-subjects design), or – Subjects were paired or matched (Matched-pairs design) l May also be called a “Repeatedmeasures” designs

Why use dependent samples? l Advantages – Fewer subjects required with a withinsubjects design

Why use dependent samples? l Advantages – Fewer subjects required with a withinsubjects design (each subject is measured twice) – Increased power through decreased variability l Disadvantages – “Carry-over” effects

Carry-over Effects Let’s say we are interested in methods to teach reading to pre-schoolers

Carry-over Effects Let’s say we are interested in methods to teach reading to pre-schoolers l We l – introduce the whole-language method, measure, – then use the phonics method, and measure l We have a repeated-measures (withinsubjects design), but

Carry-over Effects How can we remove the effects of teaching reading with the wholelanguage

Carry-over Effects How can we remove the effects of teaching reading with the wholelanguage method? l There is a “carry-over” to the next condition (phonics) condition, because the preschoolers may have learned to read l Do we expect to forget how to read once they learn to? l

Carry-over Effects Matched-pairs designs will eliminate the carry-over effect problem l Matching, though, must

Carry-over Effects Matched-pairs designs will eliminate the carry-over effect problem l Matching, though, must be done well, without respect to the obtained scores, l – matching is done before the experiment l Along a dimension that is relevant to the study

Dependent Samples t-test With pairs of scores, either the result of matching or taking

Dependent Samples t-test With pairs of scores, either the result of matching or taking multiple measurements from the same subject l Calculate the difference (D) between the two scores (maintaining positive and negative values) l Calculate the mean difference, standard error, and then t l

Dependent Samples t-test

Dependent Samples t-test

Dependent Samples t-test MD is the mean of the differences s. D is the

Dependent Samples t-test MD is the mean of the differences s. D is the standard deviation of the differences s. MD is the estimated standard error of the sampling distribution of MD Δ 0 is the difference between the population means (parameter)

Example of a Dependent Samples t-test We want to compare a new teaching method

Example of a Dependent Samples t-test We want to compare a new teaching method with a traditional method on math l 2 groups - 10 12 th grades in each, l Group “Old” receives old method, Group “New” gets new method l Subjects are matched on their IQ l

Hypothesis test of Old vs. New (teaching math) l 1. State and Check Assumptions

Hypothesis test of Old vs. New (teaching math) l 1. State and Check Assumptions – About the population l l Normally distributed? - don’t know Homogeneity of variance – we’ll check – About the sample l l Independent Random sample? – yes Dependent samples – About the sample l Interval level

Hypothesis test of Old vs. New (teaching math) l 2. Hypotheses HO : μOld

Hypothesis test of Old vs. New (teaching math) l 2. Hypotheses HO : μOld = μNew (the effectiveness of the old and new method are the same) μOld - μNew = 0 (the difference between the effectiveness of Old and New is 0) HA : μ 1 ≠ μ 2 (the effectiveness of Old and New are not equal) μ 1 - μ 2 ≠ 0 (there is a difference between the effectiveness the Old and New)

Hypothesis test of Old vs. New (teaching math) l 3. Choose test statistic –

Hypothesis test of Old vs. New (teaching math) l 3. Choose test statistic – 2 groups l dependent samples Dependent-sample t-test

Hypothesis test of Old vs. New (teaching math) l 4. Set Significance Level α

Hypothesis test of Old vs. New (teaching math) l 4. Set Significance Level α =. 05 Critical Value Non-directional Hypothesis with df = np - 1 = 10 - 1 = 9 From Table C, critical value of α /2 =. 025 tcrit = 2. 262, so we reject HO if t ≤ - 2. 262 or t ≥ 2. 262

Hypothesis test of Old vs. New (teaching math) l 5. Compute Statistic – We

Hypothesis test of Old vs. New (teaching math) l 5. Compute Statistic – We need:

Scores on the WAT (Wisconsin Achievement test) Old 78 55 95 57 60 80

Scores on the WAT (Wisconsin Achievement test) Old 78 55 95 57 60 80 50 83 90 70 New 74 45 88 65 64 75 41 68 80 64 D 4 10 7 -8 -4 5 9 15 10 6 Need: Δ MD s. MD

Scores on the WAT Old 78 55 95 57 60 80 50 83 90

Scores on the WAT Old 78 55 95 57 60 80 50 83 90 70 New 74 45 88 65 64 75 41 68 80 64 D 4 10 7 -8 -4 5 9 15 10 6 ΣD = 54 ΣD 2 = 712 MD = 5. 4 SS(D) = 420. 4 s 2 D = 46. 7111 s. D = 6. 833 np = 10 s. MD = s. D/√np = 6. 833/ √ 10 = 2. 161

Example (non- directional HA) t is greater than tcrit, reject HO

Example (non- directional HA) t is greater than tcrit, reject HO

Hypothesis test of Old vs. New (teaching math) l 6. Draw Conclusions – because

Hypothesis test of Old vs. New (teaching math) l 6. Draw Conclusions – because our t falls within the rejection region, we reject the HO, and – conclude that the old and new method differ in their effectiveness in teaching math, with the old method superior

Comparing Dependent and Independent Samples t-tests l Suppose, instead of pairing subjects based on

Comparing Dependent and Independent Samples t-tests l Suppose, instead of pairing subjects based on IQ, we compared the mean of the two groups using a 2 independent samples t-test

Scores on the WAT l Old 78 55 95 57 2219. 6 60 80

Scores on the WAT l Old 78 55 95 57 2219. 6 60 80 50 83 90 70 l ΣXO = 718 ΣX 2 O = 53772 MO = 71. 8 SS(XO) = s 2 O = 246. 622 s. O = 15. 70 n. O = 10 New 74 45 88 65 1902. 4 64 75 41 68 80 64 ΣXN = 664 ΣX 2 N = 45992 MN = 66. 4 SS(XN) = s 2 N = 211. 377 s. N = 14. 54 n. N = 10

Independent Samples t-test Cannot reject HO

Independent Samples t-test Cannot reject HO

Comparison (cont. ) By matching subjects, a priori (prior to the experiment) l the

Comparison (cont. ) By matching subjects, a priori (prior to the experiment) l the variance is reduced substantially, thereby l Allowing us to find a difference that was obscured l

Ranking When completing a Wilcoxon, ranking must be done with the smallest difference receiving

Ranking When completing a Wilcoxon, ranking must be done with the smallest difference receiving the lowest rank (1) l The biggest difference gets a rank of np l

Hypothesis Test with nonparametric alternative to t-test Nicotine patches supposedly reduces smoking l A

Hypothesis Test with nonparametric alternative to t-test Nicotine patches supposedly reduces smoking l A psychologist records the number of cigarettes each of 9 patients smokes per day without a patch and with a nicotine patch. l

OOPS!

OOPS!

Hypothesis Test l 1. State and Check assumptions – Random Sample - yes –

Hypothesis Test l 1. State and Check assumptions – Random Sample - yes – Dependent Sample (Within-subjects design) l measures come from the same person (wearing the patch, and not wearing the patch) – Number of cigarettes smokes normally distributed? - not sure – Homogeneity of Variance - nope

Hypothesis Test l 2. State Hypotheses HO : μBEFORE = μAFTER HA : μBEFORE

Hypothesis Test l 2. State Hypotheses HO : μBEFORE = μAFTER HA : μBEFORE > μAFTER

Hypothesis Test l 3. Choose Test – 2 dependent samples – H of V

Hypothesis Test l 3. Choose Test – 2 dependent samples – H of V - no Wilcoxon Tm (Wilcoxon signed-ranks test)

Hypothesis Test l 4. Set Significance level α =. 05 with np > 8,

Hypothesis Test l 4. Set Significance level α =. 05 with np > 8, the Wilcoxon Tm approaches a z directional alternative at. 05, using Table A z = 1. 65 if obtained z ≥ 1. 65, Reject HO

Hypothesis Test l 5. Compute Test Statistic

Hypothesis Test l 5. Compute Test Statistic

Wilcoxon Test l Non-parametric alternative to dependent sample t-test – Calculate the differences between

Wilcoxon Test l Non-parametric alternative to dependent sample t-test – Calculate the differences between paired scores (D) – Calculate the absolute value of each (|D|) – Rank |D| – “Sign” the ranks using the sign from D – Add all positive Ranks (T+)

Hypothesis Test l 6. Draw conclusions – Since the obtained z was greater that

Hypothesis Test l 6. Draw conclusions – Since the obtained z was greater that the critical z of +1. 65, HO is rejected, and – Conclude that the distributions are different with the before values greater than the after values

Review Procedure for calculating t Independent samples l Compute (for each group): l

Review Procedure for calculating t Independent samples l Compute (for each group): l

Calculating t - Independent Samples l To find t, the estimated standard error of

Calculating t - Independent Samples l To find t, the estimated standard error of the sampling distribution is calculated:

Computing t l The t statistic ( a family of distributions which varies with

Computing t l The t statistic ( a family of distributions which varies with df) is computed using:

Review Procedure for finding t l Dependent Samples – matched-pairs, correlated samples, repeated-measures, etc.

Review Procedure for finding t l Dependent Samples – matched-pairs, correlated samples, repeated-measures, etc. (2 samples) – Compute D (the difference between pairs of scores) l np (the number of pairs of scores l MD (the mean of the differences) l

Dependent Samples t (cont. ) – Compute s. D (the standard deviation of the

Dependent Samples t (cont. ) – Compute s. D (the standard deviation of the differences) l s. M (the estimated standard error) D l

Estimated Standard Error and t

Estimated Standard Error and t

Conceptually, what is the t statistic? In both tests, the numerator is a measure

Conceptually, what is the t statistic? In both tests, the numerator is a measure of the observed difference between scores (either unpaired or paired) l The denominator, of both tests, is an estimate of the standard deviation of the sampling distribution of M 1 - M 2 (we call it “standard error”) l

Error Terms When a sample is taken from a population, statistics computed from the

Error Terms When a sample is taken from a population, statistics computed from the sample are estimates of parameters, however, l A certain amount of variability is expected (no two sample means are exactly alike) l This variability from sample to sample is called “error” l

Error and t-tests The amount of error obtained from a sample is used to

Error and t-tests The amount of error obtained from a sample is used to estimate the “standard” error that might be expected in the situation (given n, etc. ) l This error can produce large (significant) differences between sample statistics, just by chance, because no two sample means are exactly alike l

t statistic as a ratio l Therefore, obtained difference t = ———————— difference expected

t statistic as a ratio l Therefore, obtained difference t = ———————— difference expected by chance (“error”) We will return to this type of conception again

Can’t tell the difference between independent and dependent samples? If measurements are related to

Can’t tell the difference between independent and dependent samples? If measurements are related to one another - dependent samples l You will not be told when to use independent or dependent samples ttests l You have to decide l

Example A psychologist believes that environment is more important than genetics in influencing intelligence.

Example A psychologist believes that environment is more important than genetics in influencing intelligence. l She locates 12 pairs of identical twins that have been reared apart, one twin in each pair in an enriched environment and the other in an impoverished one. l Independent or Dependent Samples? l

Another A new experimental drug, ABZ, is thought to have beneficial effects on AIDS.

Another A new experimental drug, ABZ, is thought to have beneficial effects on AIDS. l 20 AIDS patients are randomly selected and assigned to one of two conditions: ABZ for 90 days or Placebo for 90 days. l Independent or Dependent Samples? l

A third example A researcher suspects that increasing the level of lighting during the

A third example A researcher suspects that increasing the level of lighting during the winter months will increase mood. l The researcher selects 36 students, tests their mood, then replaces the light bulbs in the dorm from 75 -W to 100 -W for 1 month, then tests each student again l Independent or Dependent Sample? l