wgn replication
noah castelo
October 31, 2015
read in data, create DV and condition variables
rep<-read.csv("whogetsnudged replication.csv", sep=",", header=T)
rep$SS[rep$SSdef==1 | rep$LLdef==1 | rep$nodef==1]<-1
rep$SS[rep$SSdef==2 | rep$LLdef==2 | rep$nodef==2]<-0
rep$chosedef[rep$SSdef==1 | rep$LLdef==2]<-1
rep$chosedef[rep$SSdef==2 | rep$LLdef==1]<-0
rep$defcond[rep$SSdef>0]<-"SS"
rep$defcond[rep$nodef>0]<-"no"
rep$defcond[rep$LLdef>0]<-"LL"
rep$betac <- rep$beta - mean(rep$beta)
rep$deltac<- rep$delta - mean(rep$delta)
rep$deltacR<-(rep$deltac) * (-1)
rep$combined <- rep$betac + rep$deltacRcheck % of people who chose SS in each condition - 1 is LL, 2 is No Default, 3 is SS
prop.test(table(rep$defcond, rep$SS))##
## 3-sample test for equality of proportions without continuity
## correction
##
## data: table(rep$defcond, rep$SS)
## X-squared = 4.9797, df = 2, p-value = 0.08292
## alternative hypothesis: two.sided
## sample estimates:
## prop 1 prop 2 prop 3
## 0.5837563 0.5763547 0.4858491last time there was no difference between SS and No default - this time it’s the opposite.
both beta and delta have a significant effect on choice, but only beta interacts with default condition:
summary(glm(SS ~ betac, data=rep, family="binomial"))##
## Call:
## glm(formula = SS ~ betac, family = "binomial", data = rep)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.8787 -0.9884 -0.8208 1.1722 2.1094
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.16705 0.08677 -1.925 0.0542 .
## betac -5.58679 0.76413 -7.311 2.65e-13 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 842.91 on 611 degrees of freedom
## Residual deviance: 767.69 on 610 degrees of freedom
## AIC: 771.69
##
## Number of Fisher Scoring iterations: 4summary(glm(SS ~ deltac, data=rep, family="binomial"))##
## Call:
## glm(formula = SS ~ deltac, family = "binomial", data = rep)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.3447 -0.9313 -0.7855 1.0366 2.0569
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.16976 0.09008 -1.885 0.0595 .
## deltac 413.80450 43.96248 9.413 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 842.91 on 611 degrees of freedom
## Residual deviance: 726.63 on 610 degrees of freedom
## AIC: 730.63
##
## Number of Fisher Scoring iterations: 4summary(glm(SS ~ deltac*defcond, data=rep, family="binomial"))##
## Call:
## glm(formula = SS ~ deltac * defcond, family = "binomial", data = rep)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2774 -0.8645 -0.7423 1.0324 2.0773
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4109 0.1599 -2.569 0.01019 *
## deltac 369.7519 67.5434 5.474 4.39e-08 ***
## defcondno 0.1447 0.2234 0.647 0.51733
## defcondSS 0.5883 0.2269 2.592 0.00953 **
## deltac:defcondno 32.1379 102.4687 0.314 0.75380
## deltac:defcondSS 146.8743 113.3380 1.296 0.19501
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 842.91 on 611 degrees of freedom
## Residual deviance: 718.13 on 606 degrees of freedom
## AIC: 730.13
##
## Number of Fisher Scoring iterations: 4summary(glm(SS ~ betac*defcond, data=rep, family="binomial"))##
## Call:
## glm(formula = SS ~ betac * defcond, family = "binomial", data = rep)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.8956 -0.9591 -0.8104 1.1705 1.9623
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.3660 0.1509 -2.425 0.015307 *
## betac -4.0293 1.1008 -3.660 0.000252 ***
## defcondno 0.1375 0.2148 0.640 0.521978
## defcondSS 0.4747 0.2151 2.207 0.027306 *
## betac:defcondno -1.7019 1.7835 -0.954 0.339967
## betac:defcondSS -3.7308 1.9299 -1.933 0.053225 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 842.91 on 611 degrees of freedom
## Residual deviance: 759.21 on 606 degrees of freedom
## AIC: 771.21
##
## Number of Fisher Scoring iterations: 4plotting the effect of beta broken down by default condition - omitting the no default condition because it doesn’t differ from LL default
rep2<-subset(rep, defcond !="no")
library(visreg)
fit<-glm(SS ~ betac*defcond, data=rep2, family=binomial)
summary(fit)##
## Call:
## glm(formula = SS ~ betac * defcond, family = binomial, data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4056 -0.9611 -0.8129 1.1362 1.9623
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.3660 0.1509 -2.425 0.015307 *
## betac -4.0293 1.1008 -3.660 0.000252 ***
## defcondSS 0.4747 0.2151 2.207 0.027306 *
## betac:defcondSS -3.7308 1.9299 -1.933 0.053225 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 565.21 on 408 degrees of freedom
## Residual deviance: 505.95 on 405 degrees of freedom
## AIC: 513.95
##
## Number of Fisher Scoring iterations: 4visreg(fit, "betac", by="defcond", overlay=TRUE, partial=FALSE, xlab="present bias (beta)", scale="response", ylab="P (choosing SS)")
floodlight analysis in SPSS shows that the effect of default condition is significant everywhere below beta = 0.015. in other words, there is only a default effect for people who are relatively impatient. this isn’t quite consistent with our theory, which would predict a significant default effect only for people in the middle of the patience varialbe, but not at either extreme.
using delta as the measure of patience does give this expected pattern:
fit<-glm(SS ~ deltac*defcond, data=rep2, family=binomial)
summary(fit)##
## Call:
## glm(formula = SS ~ deltac * defcond, family = binomial, data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2774 -0.8625 -0.7300 0.9775 2.0773
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4109 0.1599 -2.569 0.01019 *
## deltac 369.7519 67.5434 5.474 4.39e-08 ***
## defcondSS 0.5883 0.2269 2.592 0.00953 **
## deltac:defcondSS 146.8743 113.3380 1.296 0.19501
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 565.21 on 408 degrees of freedom
## Residual deviance: 475.94 on 405 degrees of freedom
## AIC: 483.94
##
## Number of Fisher Scoring iterations: 4visreg(fit, "deltac", by="defcond", overlay=TRUE, partial=FALSE, xlab="discount rate (delta)", scale="response", ylab="P (choosing SS)")
the floodlight analysis here indicates a significant default effect between for the range -0.0009 < delta < 0.006, although there are only 6 people above .006.
I think this discrete analysis is most straightforward and really the most direct test of our theory:
patient people who saw the LL default and impatient people who saw the SS default saw a default that was consistent with their preferences. these people were much more likely to choose the default option than people who saw a default that was inconsistent with their preferences.
patient and impatient here is defined as 1st and 3rd quartiles - somewhat arbitrary, but if we use the floodlight points there are only 6 people who are “impatient”
rep2$consistent[(rep2$deltac<=-1.932e-03 & rep2$defcond=="LL") | (rep2$deltac>=1.541e-03 & rep2$defcond=="SS")]<-1
rep2$consistent[(rep2$deltac<=-1.932e-03 & rep2$defcond=="SS") | (rep2$deltac>=1.541e-03 & rep2$defcond=="LL")]<-0
prop.test(table(rep2$chosedef, rep2$consistent))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(rep2$chosedef, rep2$consistent)
## X-squared = 72.9514, df = 1, p-value < 2.2e-16
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## 0.4890243 0.7265866
## sample estimates:
## prop 1 prop 2
## 0.8058252 0.1980198preference strength - measured by “how easy was it / how confident were you to answer” the DEEP questions - doesn’t seem to matter. it doesn’t interact with the ‘consistent’ variable to predict whether people chose the default. our theory would predict that consistency matters more when preferences are strong.
rep2$prefstrength1<-as.numeric(rep2$prefstrength1)
rep2$prefstrength2<-as.numeric(rep2$prefstrength2)
rep2$prefstrength<-(rep2$prefstrength1 + rep2$prefstrength2)/2
summary(glm(chosedef ~ consistent*prefstrength1, data=rep2, family="binomial"))##
## Call:
## glm(formula = chosedef ~ consistent * prefstrength1, family = "binomial",
## data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.8103 -0.6820 -0.6266 0.6667 1.8571
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.5877 1.6657 -0.353 0.724
## consistent 2.0581 2.4710 0.833 0.405
## prefstrength1 -0.1881 0.3728 -0.504 0.614
## consistent:prefstrength1 0.1722 0.5482 0.314 0.753
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 282.78 on 203 degrees of freedom
## Residual deviance: 201.67 on 200 degrees of freedom
## (205 observations deleted due to missingness)
## AIC: 209.67
##
## Number of Fisher Scoring iterations: 4when we combine the data from the original study and the replication, the same pattern emerges, with beta showing an interaction with condition (default effect only among relatively impatient people), and delta showing only main effects
all<-read.csv("study 1 plus replication data.csv", header=T, sep=",")
fit3<-glm(SS ~ betac*cond, data=all, family=binomial)
summary(fit3)##
## Call:
## glm(formula = SS ~ betac * cond, family = binomial, data = all)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.5366 -1.0000 -0.7339 1.2054 2.1295
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4280 0.1087 -3.938 8.23e-05 ***
## betac -4.9734 0.8703 -5.715 1.10e-08 ***
## condSS 0.6942 0.1562 4.444 8.85e-06 ***
## betac:condSS -3.3061 1.5058 -2.196 0.0281 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1120.28 on 809 degrees of freedom
## Residual deviance: 989.26 on 806 degrees of freedom
## AIC: 997.26
##
## Number of Fisher Scoring iterations: 5visreg(fit3, "betac", by="cond", overlay=TRUE, partial=FALSE, xlab="present bias (beta)", scale="response", ylab="P (choosing SS)")
fit4<-glm(SS ~ deltac*cond, data=all, family=binomial)
summary(fit4)##
## Call:
## glm(formula = SS ~ deltac * cond, family = binomial, data = all)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2669 -0.8932 -0.6661 0.9939 2.1482
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4636 0.1147 -4.044 5.26e-05 ***
## deltac 395.9964 49.7544 7.959 1.73e-15 ***
## condSS 0.7413 0.1620 4.576 4.73e-06 ***
## deltac:condSS 92.7838 80.2682 1.156 0.248
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1120.28 on 809 degrees of freedom
## Residual deviance: 939.65 on 806 degrees of freedom
## AIC: 947.65
##
## Number of Fisher Scoring iterations: 4visreg(fit4, "deltac", by="cond", overlay=TRUE, partial=FALSE, xlab="discount rate (delta)", scale="response", ylab="P (choosing SS)")
all$combined<-(all$betac + (all$deltac*-1))
fit4<-glm(SS ~ combined*cond, data=all, family=binomial)
summary(fit4)##
## Call:
## glm(formula = SS ~ combined * cond, family = binomial, data = all)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.5495 -0.9961 -0.7339 1.2025 2.1379
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4284 0.1088 -3.937 8.24e-05 ***
## combined -4.9672 0.8624 -5.760 8.43e-09 ***
## condSS 0.6975 0.1565 4.456 8.33e-06 ***
## combined:condSS -3.2944 1.4944 -2.205 0.0275 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1120.3 on 809 degrees of freedom
## Residual deviance: 987.4 on 806 degrees of freedom
## AIC: 995.4
##
## Number of Fisher Scoring iterations: 5visreg(fit4, "combined", by="cond", overlay=TRUE, partial=FALSE, xlab="beta + delta", scale="response", ylab="P (choosing SS)")
using p(choosing default) as DV instead of p(choosing SS), which I find more intuitive
fit5<-glm(chosedefault1yes ~ deltac*cond, data=all, family=binomial)
summary(fit5)##
## Call:
## glm(formula = chosedefault1yes ~ deltac * cond, family = binomial,
## data = all)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2104 -1.0089 0.5832 0.7929 2.0272
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.5171 0.1647 3.141 0.00169 **
## deltac -425.3526 73.6216 -5.778 7.58e-09 ***
## condSS -0.1351 0.2320 -0.582 0.56028
## deltac:condSS 889.2759 114.5628 7.762 8.34e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 544.66 on 400 degrees of freedom
## Residual deviance: 461.89 on 397 degrees of freedom
## (409 observations deleted due to missingness)
## AIC: 469.89
##
## Number of Fisher Scoring iterations: 4visreg(fit5, "deltac", by="cond", overlay=TRUE, partial=FALSE, xlab="discount rate (delta)", scale="response", ylab="P (choosing default)")
discrete analysis with original study plus replication
summary(all$deltac)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -4.542e-03 -1.732e-03 -1.049e-03 2.955e-05 1.551e-03 7.140e-03all$consistent[(all$deltac<=-1.732e-03 & all$cond=="LL") | (all$deltac>=1.551e-03 & all$cond=="SS")]<-1
all$consistent[(all$deltac<=-1.732e-03 & all$cond=="SS") | (all$deltac>=1.551e-03 & all$cond=="LL")]<-0
prop.test(table(all$chosedef, all$consistent))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(all$chosedef, all$consistent)
## X-squared = 69.1601, df = 1, p-value < 2.2e-16
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## 0.4934023 0.7374355
## sample estimates:
## prop 1 prop 2
## 0.8409091 0.2254902another way to think about it - the default effect is massively reversed depending on preference consistency - there is a huge default effect in the predicted direction for people whose preferences are consistent with the default, but a huge effect in the opposite direction for people whose preferences are inconsistent - this could be seen as a behavioral measure of reactance
consistent<-subset(all, consistent==1)
inconsistent<-subset(all, consistent==0)
prop.test(table(consistent$SS, consistent$cond))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(consistent$SS, consistent$cond)
## X-squared = 83.7745, df = 1, p-value < 2.2e-16
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## 0.5448846 0.7736834
## sample estimates:
## prop 1 prop 2
## 0.815534 0.156250prop.test(table(inconsistent$SS, inconsistent$cond))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(inconsistent$SS, inconsistent$cond)
## X-squared = 62.0161, df = 1, p-value = 3.407e-15
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## -0.6815497 -0.4363074
## sample estimates:
## prop 1 prop 2
## 0.2410714 0.8000000summary(glm(SS ~ cond*consistent, data=all, family=binomial))##
## Call:
## glm(formula = SS ~ cond * consistent, family = binomial, data = all)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.8225 -0.6352 -0.5732 0.7797 1.9427
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.0349 0.2240 4.619 3.85e-06 ***
## condSS -2.5331 0.3385 -7.483 7.26e-14 ***
## consistent -2.7577 0.3588 -7.685 1.53e-14 ***
## condSS:consistent 5.7059 0.5081 11.230 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 561.42 on 405 degrees of freedom
## Residual deviance: 398.86 on 402 degrees of freedom
## (404 observations deleted due to missingness)
## AIC: 406.86
##
## Number of Fisher Scoring iterations: 4confirming that the default only exists for people in the middle range of patience
all$patient[all$deltac <=-1.732e-03]<-1
all$patient[all$deltac >= 1.551e-03]<-0
all$patient[(all$deltac > -1.732e-03) & (all$deltac < 1.551e-03)]<-3
patient<-subset(all, patient==1)
impatient<-subset(all, patient==0)
middle<-subset(all, patient==3)
prop.test(table(patient$SS, patient$cond))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(patient$SS, patient$cond)
## X-squared = 0.1653, df = 1, p-value = 0.6843
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## -0.1449320 0.2566619
## sample estimates:
## prop 1 prop 2
## 0.4970414 0.4411765prop.test(table(impatient$SS, impatient$cond))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(impatient$SS, impatient$cond)
## X-squared = 1.123, df = 1, p-value = 0.2893
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## -0.07352199 0.27928216
## sample estimates:
## prop 1 prop 2
## 0.5869565 0.4840764prop.test(table(middle$SS, middle$cond))##
## 2-sample test for equality of proportions with continuity
## correction
##
## data: table(middle$SS, middle$cond)
## X-squared = 19.7526, df = 1, p-value = 8.814e-06
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## 0.1263148 0.3259626
## sample estimates:
## prop 1 prop 2
## 0.5821596 0.3560209other analyses:
we can look at whether the default channels were more or less effective for people whose preferences were consistent vs. inconsistent with the default.
ease and reference dependence have main effects on choosing the default, but only implied endorsement interacts with “consistency”
rep2$directIE<-as.numeric(rep2$directIE)
rep2$indirectIE<-as.numeric(rep2$indirectIE)
rep2$IE<-(rep2$directIE + rep2$indirectIE)/2
rep2$refdep<-as.numeric(rep2$refdep)
rep2$ease<-as.numeric(rep2$ease)
summary(glm(SS ~ IE + refdep + ease, data=rep2, family="binomial"))##
## Call:
## glm(formula = SS ~ IE + refdep + ease, family = "binomial", data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.7594 -0.9949 -0.8816 1.1049 1.5343
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.28667 0.25272 -5.091 3.56e-07 ***
## IE -0.04270 0.14905 -0.286 0.77452
## refdep 0.28607 0.08951 3.196 0.00139 **
## ease 0.32008 0.08763 3.653 0.00026 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 565.21 on 408 degrees of freedom
## Residual deviance: 527.51 on 405 degrees of freedom
## AIC: 535.51
##
## Number of Fisher Scoring iterations: 4summary(glm(chosedef ~ IE*consistent + refdep*consistent + ease*consistent, data=rep2, family="binomial"))##
## Call:
## glm(formula = chosedef ~ IE * consistent + refdep * consistent +
## ease * consistent, family = "binomial", data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.9264 -0.5761 -0.2620 0.6003 2.2847
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.1989 0.6027 -3.649 0.000264 ***
## IE 0.3631 0.3725 0.975 0.329616
## consistent 4.2205 0.8286 5.094 3.51e-07 ***
## refdep 0.5345 0.2471 2.163 0.030549 *
## ease -0.4108 0.2609 -1.574 0.115446
## IE:consistent -1.2885 0.5741 -2.244 0.024813 *
## consistent:refdep 0.1344 0.4374 0.307 0.758670
## consistent:ease 0.2678 0.3204 0.836 0.403190
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 282.78 on 203 degrees of freedom
## Residual deviance: 184.55 on 196 degrees of freedom
## (205 observations deleted due to missingness)
## AIC: 200.55
##
## Number of Fisher Scoring iterations: 5it seems that implied endorsement (i.e. “i made my choice because i thought the researcher wanted me to select that option / because i thought most people would select that option”) has opposite effects for people whose preferences are consistent vs. inconsistent with the default. When preferences are (in)consistent, perceievd implied endorsement (increases) decreases the probability of choosing the default. not sure what to make of this.
IEcons<-glm(chosedef ~ IE*consistent, data=rep2, family="binomial")
summary(IEcons)##
## Call:
## glm(formula = chosedef ~ IE * consistent, family = "binomial",
## data = rep2)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.9095 -0.6230 -0.5558 0.5936 1.9718
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.2872 0.5617 -4.072 4.66e-05 ***
## IE 0.4976 0.2733 1.821 0.0686 .
## consistent 4.3287 0.7529 5.749 8.96e-09 ***
## IE:consistent -0.8922 0.3733 -2.390 0.0168 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 282.78 on 203 degrees of freedom
## Residual deviance: 196.36 on 200 degrees of freedom
## (205 observations deleted due to missingness)
## AIC: 204.36
##
## Number of Fisher Scoring iterations: 4visreg(IEcons, "IE", by="consistent", overlay=TRUE, partial=FALSE, xlab="Choice Influenced by Implied Endorsement", scale="response", ylab="P (choosing default)")
what about reactance? do people with default-inconsistent preferences experience greater reactance?
rep2$hong8.1<-as.numeric(rep2$hong8.1)
rep2$hong9.1<-as.numeric(rep2$hong9.1)
rep2$hong11.1<-as.numeric(rep2$hong11.1)
rep2$hong13.1<-as.numeric(rep2$hong13.1)
rep2$reactance<-(rep2$hong8.1 + rep2$hong9.1 + rep2$hong11.1 + rep2$hong13.1)/4
consistent<-subset(rep2, consistent==1)
inconsistent<-subset(rep2, consistent==0)
t.test(consistent$reactance, inconsistent$reactance)##
## Welch Two Sample t-test
##
## data: consistent$reactance and inconsistent$reactance
## t = -1.2389, df = 198.207, p-value = 0.2169
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.28962390 0.06613068
## sample estimates:
## mean of x mean of y
## 1.995050 2.106796names(rep2)
names(rep2) table(rep2\(defaultstudies) table(rep2\)defaultstudies.1)
names(rep2) rep2\(hong1<-as.numeric(rep2\)hong1) rep2\(hong2<-as.numeric(rep2\)hong2) rep2\(hong3<-as.numeric(rep2\)hong3) rep2\(hong5<-as.numeric(rep2\)hong5) rep2\(hong6<-as.numeric(rep2\)hong6) rep2\(hong7<-as.numeric(rep2\)hong7) rep2\(hong8<-as.numeric(rep2\)hong8) rep2\(hong9<-as.numeric(rep2\)hong9) rep2\(hong11<-as.numeric(rep2\)hong11) rep2\(hong12<-as.numeric(rep2\)hong12) rep2\(hong13<-as.numeric(rep2\)hong13)
rep2\(reactance<-(rep2\)hong1 + rep2\(hong2 + rep2\)hong3 + rep2\(hong5 + rep2\)hong6 + rep2\(hong7 + rep2\)hong8 + rep2\(hong9 + rep2\)hong11 + rep2\(hong12 + rep2\)hong13)/11
summary(glm(chosedef ~ consistentreactance, data=rep2, family=“binomial”)) fit3<-(glm(chosedef ~ consistentreactance, data=rep2, family=“binomial”)) visreg(fit3, “reactance”, by=“consistent”, overlay=TRUE, partial=FALSE, scale=“response”)
names(rep2)