beware of interpre ng non orthogonal contrasts
play

Beware of Interpre-ng Non-Orthogonal Contrasts > - PowerPoint PPT Presentation

Beware of Interpre-ng Non-Orthogonal Contrasts > contrasts(lev_se) > tapply(x,lev_se,mean) quad High Low Medium High 1 0 60.42472 54.49830 52.17486 Low 0 0 Medium 0 1 >


  1. Beware ¡of ¡ Interpre-ng ¡ Non-­‑Orthogonal ¡ Contrasts ¡

  2. > contrasts(lev_se) > tapply(x,lev_se,mean) quad High Low Medium High 1 0 60.42472 54.49830 52.17486 Low 0 0 Medium 0 1 > mean(tapply(x,lev_se,mean)) [1] 55.69929 > summary(lm(x~lev_se)) (Intercept) 54.498 2.019 26.991 <2e-16 *** lev_se 5.926 2.855 2.075 0.0409 * lev_sequad -2.323 2.855 -0.814 0.4181 > contrasts(lev_se) High Med High 1 0 Low -1 -1 Medium 0 1 > summary(lm(x~lev_se)) (Intercept) 55.699 1.166 47.780 < 2e-16 *** > 60.42472-55.69929 lev_seHigh 4.725 1.649 2.866 0.00521 ** [1] 4.72543 lev_seMed -3.524 1.649 -2.138 0.03534 * > contrasts(lev_se) lin quad So ¡while ¡the ¡contrasts ¡code ¡look ¡the ¡same, ¡ High 1 1 they ¡capture ¡very ¡different ¡comparisons! ¡ Low -1 1 Medium 0 -2 > summary(lm(x~lev_se)) (Intercept) 55.6993 1.1657 47.780 <2e-16 *** > (60.42472-54.49830)/2 lev_selin 2.9632 1.4277 2.075 0.0409 * [1] 2.96321 lev_sequad 1.7622 0.8243 2.138 0.0353 *

  3. p ¡ Curves ¡

  4. How ¡Are ¡ p -­‑values ¡Distributed? ¡ If ¡you ¡thought ¡you ¡had ¡a ¡weak ¡effect, ¡where ¡would ¡you ¡expect ¡ your ¡p ¡values ¡to ¡cluster? ¡

  5. Sta-s-cal ¡Power ¡ Criterion Criterion If Null Hyp. is true If Alt. Hyp. is true Power b a

  6. Distribu-on ¡of ¡obtained ¡p-­‑values? ¡

  7. Uri ¡Simonsohn’s ¡“p ¡curves” ¡

  8. How ¡Are ¡ p -­‑values ¡Distributed? ¡ Why ¡might ¡you ¡get ¡the ¡pa@ern ¡to ¡the ¡leA? ¡

  9. Credibility ¡of ¡ ReplicaDons ¡

  10. Sta-s-cal ¡Power ¡ Criterion Criterion If Null Hyp. is true If Alt. Hyp. is true Power b a

  11. Tricky ¡Consequences ¡of ¡Sta-s-cal ¡Power ¡ • The ¡best ¡way ¡to ¡dispel ¡concerns ¡is ¡to ¡show ¡ mulDple ¡replicaDons, ¡right? ¡ • Say ¡that ¡you ¡show ¡5 ¡studies ¡that ¡all ¡replicate ¡ the ¡effect ¡ • Say ¡that ¡your ¡effecDve ¡power ¡is ¡60% ¡ • The ¡probability ¡of ¡all ¡5 ¡tests ¡being ¡significant ¡is ¡ 7.8% ¡(.60^5) ¡ • ReplicaDons ¡can ¡hurt ¡your ¡credibility!!! ¡

  12. Advanced ¡Topics: ¡ ¡Within-­‑Subject ¡Media-on ¡& ¡ Moderated ¡Media-on ¡ 12 ¡

  13. Within-­‑Subject ¡Media-on ¡ Time ¡1: ¡ Time ¡2: ¡ OpDmism1 ¡ OpDmism2 ¡ IntenDons1 ¡ IntenDons2 ¡ Cras ash h Model: ¡ ¡ Crash à ¡OpDmism ¡ à ¡IntenDons ¡ ¡ Imagine ¡that ¡a ¡researcher ¡is ¡curious ¡about ¡changes ¡in ¡couples’ ¡intenDon ¡to ¡ conceive ¡before ¡and ¡aAer ¡the ¡financial ¡crash. ¡Furthermore, ¡he ¡believes ¡that ¡ any ¡change ¡he ¡finds ¡is ¡mediated ¡by ¡changes ¡in ¡general ¡opDmism. ¡ ¡ Other ¡Examples: ¡ (Husband,Wife) ¡ à ¡Sense ¡of ¡IdenDty ¡ à ¡SaDsfacDon ¡with ¡Marriage ¡ Rhyming ¡ à ¡Fluency ¡ à ¡Familiarity à Truth/Buying ¡IntenDons ¡ ¡

  14. Med a b IV DV IV DV c c’

  15. Within-­‑Subject ¡Media-on ¡ Judd, ¡C.M., ¡Kenny, ¡D.A., ¡& ¡McClelland, ¡G.H. ¡(2001). ¡EsDmaDng ¡and ¡tesDng ¡mediaDon ¡and ¡ Model: ¡ ¡ Crash à ¡OpDmism ¡ à ¡IntenDons ¡ ¡ moderaDon ¡in ¡within-­‑subject ¡designs. ¡ Psychological ¡Methods, ¡6 , ¡115-­‑134. ¡ Remember ¡that ¡a ¡within-­‑subject ¡or ¡paired-­‑sample ¡ t -­‑test ¡is ¡equivalent ¡to ¡ tesDng ¡a ¡one-­‑sample ¡ t -­‑test ¡on ¡a ¡difference ¡score ¡computed ¡for ¡each ¡ respondent, ¡i.e. ¡ D intenDons . ¡Here ¡it ¡is ¡the ¡test ¡of ¡ c . ¡Similarly, ¡it ¡is ¡fairly ¡intuiDve ¡ that ¡a ¡test ¡of ¡ D opDmism ¡is ¡a ¡test ¡of ¡ a . ¡The ¡next ¡step ¡it ¡to ¡predict ¡ D intenDons ¡with ¡ D opDmism . ¡The ¡slope ¡of ¡ D opDmism ¡is ¡a ¡test ¡of ¡ b. ¡ MediaDon ¡is ¡tested ¡by ¡looking ¡at ¡ the ¡significance ¡of ¡the ¡intercept. ¡ ¡ (1) ¡ D DV ¡ ≠ ¡0 ¡? ¡ ¡ ¡ ¡ ¡ ó ¡c ¡ (2) ¡ D Med ¡ ≠ ¡0 ¡? ¡ ¡ ¡ ¡ ¡ ó ¡ a ¡ (3) ¡Does ¡ D Med ¡ predict ¡ D DV ¡? ¡ ¡ ¡ ó ¡ b ¡ (4) ¡Intercept ¡ ≠ ¡0 ¡ ¡or ¡compared ¡to ¡(1) ¡? ¡ ó ¡ c’ ¡ It ¡turns ¡out ¡that ¡it ¡makes ¡sense ¡to ¡also ¡include ¡the ¡sum ¡of ¡mediators, ¡ S med , ¡ centered, ¡in ¡the ¡regression ¡equaDon. ¡This ¡tests ¡moderaDon, ¡i.e., ¡ whether ¡the ¡relaDonship ¡between ¡intenDons ¡and ¡opDmism ¡is ¡constant ¡ at ¡Time ¡1 ¡and ¡Time ¡2. ¡(See ¡Judd ¡et ¡al., ¡2001). ¡

  16. Within-­‑Subject ¡Media-on: ¡Example ¡1 ¡ 80 80 75 75 70 70 intentions1 intentions2 65 65 60 60 55 55 50 50 0 0 2 2 4 4 6 6 8 8 optimism1 optimism2

  17. Within-­‑Subject ¡Media-on: ¡Example ¡1 ¡ IntenDons1 ¡ IntenDons2 ¡ *3 ¡ *3 ¡ -­‑2 ¡ optimism1<-rnorm(100,5,1) OpDmism1 ¡ OpDmism2 ¡ optimism2<-optimism1-2+rnorm(100,0,1) intentions1<-50+3*optimism1+rnorm(100,0,5) intentions2<-50+3*optimism2+rnorm(100,0,5) cbind(optimism1,optimism2,intentions1,intentions2)->crashd round(cor(crashd),2) optimism1 optimism2 intentions1 intentions2 optimism1 1.00 0.65 0.48 0.36 optimism2 0.65 1.00 0.40 0.59 intentions1 0.48 0.40 1.00 0.18 intentions2 0.36 0.59 0.18 1.00 diff_optimism<-optimism2-optimism1 diff_intentions<-intentions2-intentions1 csum_optimism<-scale(optimism2+optimism1,scale=F) # By using scale = F we center it without dividing it by the s.d.

  18. Within-­‑Subject ¡Media-on: ¡Example ¡1 ¡ > t.test(optimism1,optimism2,paired=T) Paired t-test t = 20.565 , df = 99, p-value < 2.2e-16 alternative hypothesis: true difference in means is not equal to 0 > t.test(diff_optimism) One Sample t-test t = -20.565 , df = 99, p-value < 2.2e-16 alternative hypothesis: true mean is not equal to 0 > summary(lm(diff_optimism~1)) Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -2.03701 0.09905 -20.57 <2e-16 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.9905 on 99 degrees of freedom Many ¡ways ¡to ¡skin ¡a ¡cat: ¡The ¡point ¡is ¡to ¡show ¡you ¡the ¡use ¡of ¡lm(X~1), ¡the ¡ model ¡that ¡only ¡includes ¡the ¡intercept. ¡

  19. Within-­‑Subject ¡Media-on: ¡Example ¡1 ¡ > summary(lm(diff_intentions~1)) Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -5.5870 0.7716 -7.241 9.76e-11 *** > summary(lm(diff_intentions~diff_optimism+csum_optimism)) Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.5351 1.8189 -0.294 0.76925 diff_optimism 2.4801 0.8150 3.043 0.00301 ** csum_optimism -0.2316 0.3972 -0.583 0.56121 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 7.435 on 97 degrees of freedom Multiple R-squared: 0.09038, Adjusted R-squared: 0.07163 F-statistic: 4.819 on 2 and 97 DF, p-value: 0.01011 We ¡would ¡conclude ¡that ¡opDmism ¡mediates ¡the ¡effect ¡of ¡the ¡crash ¡on ¡ intenDons, ¡and ¡that ¡the ¡relaDonship ¡between ¡opDmism ¡and ¡intenDons ¡is ¡ similar ¡in ¡both ¡groups. ¡

  20. Within-­‑Subject ¡Media-on: ¡Example ¡2 ¡ Evaluate ¡30 ¡aphorisms ¡for ¡truth ¡and ¡familiarity: ¡ Non-­‑Rhyming ¡ Rhyming ¡ When ¡good ¡cheer ¡is ¡absent, ¡friends ¡ When ¡good ¡cheer ¡is ¡lacking, ¡friends ¡will ¡ will ¡go ¡packing. ¡ go ¡packing. ¡ A ¡person ¡without ¡reason ¡is ¡a ¡deer ¡ A ¡person ¡without ¡reason ¡is ¡a ¡deer ¡in ¡ being ¡hunted. ¡(etc.) ¡ season. ¡(etc.) ¡ • ¡Within ¡individuals ¡ 4 Average ¡r (truth,familiarity) =.30 ¡ Rhyming Skurnilk, ¡Monin, ¡& ¡ ¡Dewan ¡(in ¡prep.) ¡ 3.9 Non-rhyming 95%CI ¡= ¡[.25, ¡.34] ¡ 3.8 3.7 3.6 • ¡D familiarity ¡ à ¡D truth ¡: ¡ 1-7 ratings 3.5 ¡β=.245, ¡t(150)=3.15, ¡p=.002 ¡ 3.4 ¡intercept ¡t=1.2, ¡n.s. ¡ 3.3 3.2 • ¡D truth ¡ à ¡D familiarity ¡: ¡ 3.1 ¡β=.245, ¡t(150)=3.15, ¡p=.002 ¡ 3 intercept ¡t=4.41, ¡p<.0005 ¡ Truth Familiarity ¡

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend