creating the “test-retest reliability measure” - since the DEEP estimates are just single numbers the most intuitive measure is just the difference between the new and old estimates
then testing whether this measure of preference certainty interacts with default condition to predict choice of SS
r$betadiff<-abs(r$beta.x - r$beta.y)
summary(glm(SS~cond1SS2LL3NO*betadiff, data=r, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO * betadiff, family = binomial,
## data = r)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.9460 -1.0438 -0.9639 1.2561 1.4148
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.4015 0.2812 -1.428 0.1533
## cond1SS2LL3NO -0.0471 0.1339 -0.352 0.7250
## betadiff 5.1525 2.3315 2.210 0.0271 *
## cond1SS2LL3NO:betadiff -0.6720 1.0688 -0.629 0.5295
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 887.01 on 642 degrees of freedom
## Residual deviance: 864.12 on 639 degrees of freedom
## AIC: 872.12
##
## Number of Fisher Scoring iterations: 4
r$deltadiff<-abs(r$delta.x - r$delta.y)
summary(glm(SS~cond1SS2LL3NO*deltadiff, data=r, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO * deltadiff, family = binomial,
## data = r)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.7231 -1.0340 -0.9399 1.2501 1.4457
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.2548 0.2791 -0.913 0.361
## cond1SS2LL3NO -0.1190 0.1337 -0.890 0.374
## deltadiff 223.3119 145.0674 1.539 0.124
## cond1SS2LL3NO:deltadiff 9.0705 67.2151 0.135 0.893
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 887.01 on 642 degrees of freedom
## Residual deviance: 864.61 on 639 degrees of freedom
## AIC: 872.61
##
## Number of Fisher Scoring iterations: 4
neither reliability measure (beta nor delta) interacts with condition the original and latest estimates of beta and delta are moderately correlated, but the original estimate is a stronger predictor of choice (the choice occured 1 week after the original measure vs. several months later) the original estimate of beta also interact with condition - the latest one does not
cor.test(r$beta.x, r$beta.y)
##
## Pearson's product-moment correlation
##
## data: r$beta.x and r$beta.y
## t = 19.484, df = 641, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
## 0.5589122 0.6562501
## sample estimates:
## cor
## 0.6098762
cor.test(r$delta.x, r$delta.y)
##
## Pearson's product-moment correlation
##
## data: r$delta.x and r$delta.y
## t = 21.046, df = 641, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
## 0.5911448 0.6828170
## sample estimates:
## cor
## 0.6392464
#original measure of beta
summary(glm(SS~cond1SS2LL3NO*beta.y, data=r, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO * beta.y, family = binomial,
## data = r)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.5718 -0.9970 -0.7367 1.1728 2.0241
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 12.097 2.421 4.996 5.84e-07 ***
## cond1SS2LL3NO -3.042 1.019 -2.984 0.00284 **
## beta.y -12.098 2.384 -5.075 3.87e-07 ***
## cond1SS2LL3NO:beta.y 2.952 1.002 2.947 0.00321 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 887.01 on 642 degrees of freedom
## Residual deviance: 798.69 on 639 degrees of freedom
## AIC: 806.69
##
## Number of Fisher Scoring iterations: 4
#newer measure of beta
summary(glm(SS~cond1SS2LL3NO*beta.x, data=r, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO * beta.x, family = binomial,
## data = r)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.419 -1.000 -0.661 1.150 2.147
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 6.6263 2.0220 3.277 0.00105 **
## cond1SS2LL3NO -0.3621 0.9589 -0.378 0.70570
## beta.x -6.5756 1.9950 -3.296 0.00098 ***
## cond1SS2LL3NO:beta.x 0.2536 0.9480 0.267 0.78909
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 887.01 on 642 degrees of freedom
## Residual deviance: 796.19 on 639 degrees of freedom
## AIC: 804.19
##
## Number of Fisher Scoring iterations: 4
the new measure of preference certainty (alpha=.91) is very skewed - people are very certain about their DEEP preferences
hist(r$certain)
summary(r$certain)
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## 2.600 6.000 6.600 6.292 7.000 7.000 9
nevertheless, there’s an almost-marginally significant interaction between preference certainty and default condition
summary(glm(SS~cond1SS2LL3NO*certain, data=r, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO * certain, family = binomial,
## data = r)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.886 -1.078 -1.023 1.269 1.340
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 4.0832 1.7634 2.315 0.0206 *
## cond1SS2LL3NO -1.3256 0.8057 -1.645 0.0999 .
## certain -0.6416 0.2780 -2.308 0.0210 *
## cond1SS2LL3NO:certain 0.1942 0.1275 1.523 0.1277
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 874.96 on 633 degrees of freedom
## Residual deviance: 865.25 on 630 degrees of freedom
## (9 observations deleted due to missingness)
## AIC: 873.25
##
## Number of Fisher Scoring iterations: 4
a spotlight analysis (in SPSS) shows that the effect of condition on choice of SS is strongest when preference certainty is lowest, and becomes marginally significant when certain is less than 5.3 on the 7 point scale (N=65, so only about 20 per default condition)
certain<-subset(r, certain>5.3)
un<-subset(r, certain<=5.3)
summary(glm(SS~cond1SS2LL3NO, data=un, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO, family = binomial, data = un)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.6770 -1.1854 0.7499 0.9461 1.1694
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.6781 0.7312 2.295 0.0217 *
## cond1SS2LL3NO -0.5531 0.3223 -1.716 0.0862 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 85.611 on 64 degrees of freedom
## Residual deviance: 82.540 on 63 degrees of freedom
## AIC: 86.54
##
## Number of Fisher Scoring iterations: 4
summary(glm(SS~cond1SS2LL3NO, data=certain, family=binomial))
##
## Call:
## glm(formula = SS ~ cond1SS2LL3NO, family = binomial, data = certain)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.103 -1.077 -1.050 1.282 1.310
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.11332 0.22016 -0.515 0.607
## cond1SS2LL3NO -0.06432 0.10621 -0.606 0.545
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 780.89 on 568 degrees of freedom
## Residual deviance: 780.53 on 567 degrees of freedom
## AIC: 784.53
##
## Number of Fisher Scoring iterations: 3