150 likes | 328 Views
ANCOV. Confounded Predictors. Confound.sas. data confound; input gender courses aptitude pair apt1 apt2 diff; interaction = gender*courses; cards ; …………….. Data here …………. proc corr ; var gender courses aptitude ; Courses = number of literature courses taken Aptitude = verbal aptitude
E N D
ANCOV Confounded Predictors
Confound.sas data confound; input gender courses aptitude pair apt1 apt2 diff; interaction = gender*courses; cards; …………….. Data here …………. proccorr; var gender courses aptitude; Courses = number of literature courses taken Aptitude = verbal aptitude Gender 1 = female, 2 = male Contrived data.
Significant gender difference on number of courses and reading aptitude. • Significant correlation between verbal aptitude and number of literature courses taken.
Courses: Independent Samples t • procttest; class gender; var courses aptitude; • Women took significantly more courses than did men, t(32) = 4.05, p < .001
Aptitude • Women had significantly greater verbal aptitude, t(32) = 2.68, p = .012
Match Subjects on # Courses • Match subjects on number of lit courses taken. • Conduct matched-pairs t test comparing the two genders • Data from many of the highest scoring women dropped due to lack of men that score so high. • Data from many of the lowest scoring men dropped due to lack of women that score so low.
Matched Pairs t • procmeans mean stddev n t prt;var apt1 apt2 diff; • Now the men score significantly higher than do the women, t(9) = 5.93, p < .001.
ANCOV Using All Data • After showing that the Gender x Aptitude interaction is not significant, do ANCOV procglm; class gender; model aptitude = courses gender / ss1; means gender; lsmeans gender;
Least Squares Means • After holding constant the effect of number of literature courses taken, men have verbal aptitude that is significantly greater than that of women, F(1, 31) = 18.26, p < .001
Weights.sas procformat; value gen 1='Female' 2='Male'; data weights; input gender height weight; interaction = gender*height; format gender gen. ; cards; 2 70 172 2 74 130 …………….. Rest of Data ……………
Zero-Order Corrs • proccorr; var gender height weight; • These are from PDS data, subjects are grad students.
Heights • procttest; class gender; var height weight; • Men are significantly taller than women, t(47) = 8.00, p < .001. • Men are 5.68 inches taller than women.
Weights • Men weigh significantly more than do women, t(47) = 8.76, p < .001. • The mean difference is 40.4 pounds.
ANCOV procglm; class gender; model weight= height gender / ss1; means gender; lsmeans gender; • There was no significant Gender x Height interaction. • If men and women did not differ on height, would they differ on weight?
Men weigh significantly more than women even when controlling for height. • Do note that the difference has been reduced from 40.4 pounds to 35.24 pounds