J. T. Stocks, Ph.D. Group Designs and Statistical Hypothesis Testing Spurious Relationships Age Shoe Size Causal (Functional) Relationships Test Score Intervening Variable Relationship Two variables are said to be functionally related if all of the following three criteria are met. !Relationship: there is a relationship between the two Seen as threat variables. !Asymmetry: change in the “cause” variable results in change in the “effect” variable, but not vice versa. !Nonspuriousness: change in the “cause” variable results in change in the “effect” variable regardless of the actions of other variables. Criticism Aggression Symmetry and Asymmetry Causal (Functional) Relationships Symmetrical Relationship Two variables are said to be functionally related if all of the following three criteria are met. !Relationship: there is a relationship between the two A B variables. Asymmetrical Relationship A B !Asymmetry: change in the “cause” variable results in change in the “effect” variable, but not vice versa. !Nonspuriousness: change in the “cause” variable results in change in the “effect” variable regardless of the actions of other variables. 1 J. T. Stocks, Ph.D. Dependent and Independent Variables Threats To Validity !Internal validity refers to the correctness !Independent: It is the causal variable; it is the variable that is manipulated by the researcher. !Dependent: It is the effect variable; it is the outcome variable. of a conclusion about functional relationships between variables. A conclusion is internally valid to the extent to which it meets the criteria for causality outlined previously. !External validity refers to the correctness of a generalization of a relationship beyond the study sample. Hypotheses !Prediction About Relationship Between Independent and Dependent Variable !Empirical, Not Normative !Testable “Bad” Hypotheses !Adults who have been sexually abused as children may or may not remember the abuse. !As complexity increases, formalism increases. !Additional money ought to be allocated for child care so that parents required to work under the work provisions of the Temporary Assistance to Needy Families Act will not have to leave their children unsupervised. Internal Validity - History An unanalyzed event (extraneous variable) that may affect the dependent variable. It is the “confounding” variable that may lead to either the conclusion that there is a relationship between two variables when there is none or that there is no relationship when in fact there is one.. Internal Validity – Maturation The passage of time and the natural changes that take place over time. In absence of control, these changes may be mistaken for the effect of an independent variable. 2 J. T. Stocks, Ph.D. Internal Validity - Testing Internal Validity - Mortality The effects of being observed or measured. More specifically, it refers to the effect upon the behavior of the individual being observed. Internal Validity - Instrumentation Inaccuracy in the evaluation or measurement process. Use of measurement techniques with low reliability would fall under this rubric. Also included would be changes in the measurement technique over the course of a study. Internal Validity - Regression This refers to the tendency of extreme events to be followed by less extreme events. Very high or very low scores tend to regress, or move toward the mean score. Mortality refers to the fact the some individuals may drop out of a study and their absence may have a significant effect on the study’s outcome. Internal Validity - Selection Pre-existing differences between experimental and control groups may be mistaken for the effect of an independent variable. Internal Validity Interactions with Selection This occurs when any of the previous effects interact with selection. For example: !A history-selection interaction would occur when one group experiences certain events that the other does not. !A regression-selection interaction would occur when one group was more extreme on measures of the dependent variable than the other. !A mortality-selection threat would occur when there were differences between groups with respect to dropout rates. 3 J. T. Stocks, Ph.D. Internal Validity Diffusion of Treatment This refers to a situation where there is communication between members of the treatment and comparison groups leading to some degree of duplication of treatment between the groups. Internal Validity Resentful Demoralization This occurs when the members of the group receiving the perceived “less desirable” treatment become demoralized and simply “give up.” Internal Validity Compensatory Equalization Requirements for Generalization This threat also involves duplication of treatments between groups by an external agent (e.g., administrator, therapist). This is usually based in the belief that the experimental intervention is more effective than the comparison intervention. !Consistency: There is a consistency inherent in the situation under discussion that permits generalization from observed events to unobserved events !Statistical Validity: The samples discussed in the premises have been randomly drawn thus allowing for exact evaluation of the inductive probability of the argument. Internal Validity Compensatory Rivalry This occurs when the members of the group receiving the perceived “less desirable” treatment are motivated to exert themselves more strongly so as to outperform the “privileged” group. Standard Error of the Mean Population The Standard Error of the Mean is the standard deviation of the distribution of sample means. It is computed as σX = SEM = σX / √n where n is the sample size 4 J. T. Stocks, Ph.D. Origins of the t Distribution William S. Gossett - writing under the name of “Student” introduced the t-statistic and its sampling distribution - the t distribution. Gossett was a brewer when engaged in his day job. Two Distributions The z Distribution X - µX z = ———— σX Only one z distribution The t distribution X - µX t = ———— sX Potentially an infinite array of t distributions. Different distribution for each degrees of freedom. Standard Error of the Mean Estimated from Sample Data The Standard Error of the Mean is estimated from sample data using the sample standard deviation. It is computed as sX = SEM = sX / √n where n is the size of the sample used to estimate the standard deviation. Confidence Intervals Sets upper and lower limits for µX at a given level of probability !Upper Limit = X + t(αα/2tail,df) • sX !Lower Limit = X - t(αα/2tail,df) • sX A 95% confidence interval = (1 - α) • 100% Thus, t(αα/2tail,df) would be t(.05/2tail,df) df = n-1 Compute Upper and Lower Limits X 17 5 11 sX 10 10 7 n 25 9 16 Null Hypothesis and Causality Criteria for Causality !Relationship !Asymmetry (Temporality) !Nonspuriousness 5 J. T. Stocks, Ph.D. Null Hypothesis and Sampling ErrorError-1 H0: µ1 = µ2 H0: P(A) = P(~A) Assumptions for All Hypothesis Tests !Randomness !Independence The Null Hypothesis (H0) is the "no difference" or "no relationship" hypothesis. Null Hypothesis and Sampling ErrorError-2 H0: µ1 - µ2 = 0 H0: P(A) – P(~A) = 0 H0: µ1 - µ2 = 0 The Map Is Not The Territory - 1 Test: X1 - X2 Decision Rules and α and p p refers to the probability that a relationship or a difference of a certain size would be seen in a sample if the H0 were true. To reject H0, p must be less than or equal to α. !Rejecting the H0: we believe that it is likely that the relationship in the sample is generalizable to the population. !Not rejecting the H0: we do not believe we have sufficient evidence to draw inferences about the population. 6 J. T. Stocks, Ph.D. OneOne- versus TwoTwo-Tailed Tests Effect Size |µ1 - µ2| d = ———— σ g = |P(A) - 0.50| H0: µ1 - µ2 > 0 Directional Nondirectional H0: µ1 - µ2 = 0 Program Comparison Change in Corporal Punishment 10 Small Effects !Small Effect Size: d = 0.2. g = 0.05. Behavior Management Mean = -4.36 Anger Management Mean = +3.19 8 6 Behavior Anger 4 2 This is the equivalent of a PVE of approximately 0.01. It is approximately the effect size for the average difference in height (i.e., 0.5 inches and s = 2.1) between 15- and 16-year-old girls (Cohen, 1988). 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 The Alternative Hypothesis (Experimental, Research) Medium Effects !Medium Effect Size: d = 0.5. g = 0.15. H1: µ1 - µ2 ≥ d H1: P(A) – P(~A) ≥ g This is the equivalent of a PVE of approximately 0.06. It is approximately the effect size for the average difference in height (i.e., 1.0 inches and s = 2.0) between 14- and 18-year-old girls (Cohen, 1988). 7 J. T. Stocks, Ph.D. Large Effects Type II Error and β Level !Large Effect Size: d = 0.8. g = 0.25. This is the equivalent of a PVE of approximately 0.14. This is the same effect size as the average difference in height for 13- and 18-year-old girls (Cohen, 1988) The Map Is Not the Territory - 2 Decision Rules - 1 H0: The defendant is innocent. H1: The defendant is guilty of a specific crime. !The defense and prosecution present evidence. !The jury evaluates the evidence under the presumption of innocence rule. !The probability of the evidence under the presumption of innocence is evaluated according to the criterion of “reasonable doubt.” Decision Rules - 2 H0: There is no relationship between X & Y. H1: There is a specific relationship between X & Y. H1: µ1 - µ2 ≥ 10 !The researcher collects sample data. !A statistical procedure is used to determine the probability (p) that the sample data could have occurred if the H0 were true. !The value of p is compared with the α level. If p is less than or equal to α , then the researcher rejects the H0. 8 J. T. Stocks, Ph.D. Effect of Sample Size on Power n=9 α = .05 σ = 60 minutes d = 0.5 SEM = σ/√n H0: µpost – µpre = µchange = 0 !Concern "Nursing home residents participation in rehabilitation is low !Results of Intervention "Increase of µchange = 30 minutes d = µchange/SEM = _____ p = _____ Effect of Sample Size on Power n = 25 α = .05 d = 0.5 SEM = σ/√n σ = 60 minutes H0: µpost – µpre = µchange = 0 !Concern "Nursing home residents participation in rehabilitation is low !Results of Intervention "Increase of µchange = 30 minutes d = µchange/SEM = _____ p = _____ 9