# taskus study 1: initial N = 239 - final n: 199 (83.26%) # taskus study 2: initial N = 289 ```r completed condition N perc <int> <char> <int> <num> 1: 0 50 17.30 2: 1 control 134 46.37 3: 1 treatment 105 36.33 ``` covariates ~ condition - only AOT differs between condition - see also [supplement table 6](https://docs.google.com/document/d/1UWP7SFx_M8P6WMenbVYUeOiBcR6_31duEeArR1gJ_xY/edit) ```r > feols(c(ageZ, genderZ, aotZ, covidconcernZ, sesZ, attentionZ, educationZ, godZ, hireZ) ~ condition, data = d11) Standard-errors: IID Dep. var.: ageZ Estimate Std. Error t value Pr(>|t|) (Intercept) 0.068970 0.096966 0.711286 0.47762 conditiontreatment -0.011517 0.144754 -0.079564 0.93665 --- Dep. var.: genderZ Estimate Std. Error t value Pr(>|t|) (Intercept) 0.133870 0.100772 1.32845 0.18567 conditiontreatment -0.212957 0.146505 -1.45358 0.14777 --- Dep. var.: aotZ (maybe because treatment reminds them of accuracy) Estimate Std. Error t value Pr(>|t|) (Intercept) -0.124831 0.086310 -1.44632 0.1494087 conditiontreatment 0.361456 0.130216 2.77583 0.0059457 ** --- Dep. var.: covidconcernZ Estimate Std. Error t value Pr(>|t|) (Intercept) -0.037470 0.086967 -0.430852 0.66697 conditiontreatment 0.111966 0.131208 0.853347 0.39433 --- Dep. var.: sesZ Estimate Std. Error t value Pr(>|t|) (Intercept) 0.094717 0.084101 1.12623 0.26122 conditiontreatment -0.146973 0.126690 -1.16010 0.24719 --- Dep. var.: attentionZ Estimate Std. Error t value Pr(>|t|) (Intercept) -0.047982 0.089313 -0.537237 0.59164 conditiontreatment 0.104756 0.140902 0.743467 0.45799 --- Dep. var.: educationZ Estimate Std. Error t value Pr(>|t|) (Intercept) -0.034824 0.088677 -0.392706 0.69489 conditiontreatment 0.004338 0.133865 0.032405 0.97418 --- Dep. var.: godZ Estimate Std. Error t value Pr(>|t|) (Intercept) 0.038864 0.090648 0.428738 0.66851 conditiontreatment -0.166919 0.136841 -1.219801 0.22376 --- Dep. var.: hireZ Estimate Std. Error t value Pr(>|t|) (Intercept) 0.05875 0.086011 0.683052 0.49524 conditiontreatment -0.08133 0.129765 -0.626748 0.53143 ``` ```r causal forest ATE: b = -0.13 [-0.26, -0.01], p = .014 ATT: b = -0.14 [-0.27, -0.01] ``` # lucid studies ```r # qualtrics raw: d00 > d00[, n_distinct(id), keyby = .(country)] Key: <country> country V1 <char> <int> 1: IN 2949 2: PN 2486 > d00[, n_distinct(id), keyby = .(country, condition)] Key: <country, condition> country condition V1 <char> <char> <int> 1: IN control 1467 2: IN treatment 1482 3: PN control 1246 4: PN treatment 1240 # passed 2 bot checks: d01 > d01[, n_distinct(id), keyby = .(country)] Key: <country> country V1 <char> <int> 1: IN 1908 2: PN 1795 > d01[, n_distinct(id), keyby = .(country, condition)] Key: <country, condition> country condition V1 <char> <char> <int> 1: IN control 958 2: IN treatment 950 3: PN control 892 4: PN treatment 903 # chi-square test for attrition country statistic p <char> <num> <num> 1: IN 0.4143326 0.5197786 2: PN 0.4117092 0.5211033 # passed first dogcat screener > d02[, n_distinct(id), keyby = .(country)] Key: <country> country V1 <char> <int> 1: IN 1346 2: PN 1583 > d02[, n_distinct(id), keyby = .(country, condition)] Key: <country, condition> country condition V1 <char> <char> <int> 1: IN control 680 2: IN treatment 666 3: PN control 786 4: PN treatment 797 # chi-square test for attrition country statistic p <char> <num> <num> 1: IN 0.5383571 0.4631144 2: PN 0.3321915 0.5643715 # passed second screener screener_3 Key: <country> country V1 <char> <int> 1: IN 1065 2: PN 1300 > d03[, n_distinct(id), keyby = .(country, condition)] Key: <country, condition> country condition V1 <char> <char> <int> 1: IN control 548 2: IN treatment 517 3: PN control 660 4: PN treatment 640 # chi-square test for attrition country statistic p <char> <num> <num> 1: IN 1.8436618 0.1745224 2: PN 0.4057094 0.5241553 # final completed/reported > (t11111 <- d2[, n_distinct(id), keyby = .(country)]) Key: <country> country V1 <char> <int> 1: IN 997 2: PN 1184 > (t22222 <- d2[, n_distinct(id), keyby = .(country, condition)]) Key: <country, condition> country condition V1 <char> <char> <int> 1: IN control 512 2: IN treatment 485 3: PN control 610 4: PN treatment 574 # chi-square test for attrition country statistic p <char> <num> <num> 1: IN 1.462943 0.2264624 2: PN 1.666086 0.1967836 ``` Treatment and control group differ on these covariates, only in India - belief in god (lower in treatment) - AOT (higher in treatment) ```r # india > feols(c(gender, god, age, education, income, covid_concern, conspiracy, aot) ~ condition, data = d3[country == "IN"]) Standard-errors: IID Dep. var.: gender Estimate Std. Error t value Pr(>|t|) (Intercept) 1.439453 0.021939 65.612722 < 2.2e-16 *** conditiontreatment -0.006463 0.031455 -0.205484 0.83724 --- Dep. var.: god Estimate Std. Error t value Pr(>|t|) (Intercept) 6.974609 0.079105 88.16955 < 2.2e-16 *** conditiontreatment -0.261207 0.113417 -2.30307 0.02148 * --- Dep. var.: age Estimate Std. Error t value Pr(>|t|) (Intercept) 35.369141 0.572583 61.771228 < 2.2e-16 *** conditiontreatment 0.259725 0.820947 0.316373 0.75179 --- Dep. var.: education Estimate Std. Error t value Pr(>|t|) (Intercept) 5.275391 0.057919 91.082508 < 2.2e-16 *** conditiontreatment -0.071267 0.083042 -0.858206 0.39099 --- Dep. var.: income Estimate Std. Error t value Pr(>|t|) (Intercept) 4.177734 0.136380 30.633106 < 2.2e-16 *** conditiontreatment -0.037528 0.195536 -0.191925 0.84784 --- Dep. var.: covid_concern Estimate Std. Error t value Pr(>|t|) (Intercept) 83.871094 0.933554 89.840685 < 2.2e-16 *** conditiontreatment -0.963877 1.338493 -0.720121 0.47162 --- Dep. var.: conspiracy Estimate Std. Error t value Pr(>|t|) (Intercept) 77.99238 0.591762 131.79684 < 2.2e-16 *** conditiontreatment -1.00393 0.848445 -1.18326 0.23699 --- Dep. var.: aot Estimate Std. Error t value Pr(>|t|) (Intercept) 2.636719 0.049059 53.74600 < 2.2e-16 *** conditiontreatment 0.192010 0.070339 2.72979 0.0064494 ** ``` ```r # philippines > feols(c(gender, god, age, education, income, covid_concern, conspiracy, aot) ~ condition, data = d3[country == "PN"]) Standard-errors: IID Dep. var.: gender Estimate Std. Error t value Pr(>|t|) (Intercept) 1.513115 0.020258 74.692660 < 2.2e-16 *** conditiontreatment -0.011373 0.029095 -0.390882 0.69595 --- Dep. var.: god Estimate Std. Error t value Pr(>|t|) (Intercept) 7.372131 0.062850 117.297675 < 2.2e-16 *** conditiontreatment -0.084675 0.090266 -0.938059 0.34841 --- Dep. var.: age Estimate Std. Error t value Pr(>|t|) (Intercept) 37.177049 0.557061 66.73784 < 2.2e-16 *** conditiontreatment 0.073822 0.800060 0.09227 0.9265 --- Dep. var.: education Estimate Std. Error t value Pr(>|t|) (Intercept) 4.301639 0.050072 85.90952 < 2.2e-16 *** conditiontreatment 0.093831 0.071914 1.30477 0.19223 --- Dep. var.: income Estimate Std. Error t value Pr(>|t|) (Intercept) 2.872131 0.104869 27.38786 < 2.2e-16 *** conditiontreatment -0.070737 0.150614 -0.46966 0.63868 --- Dep. var.: covid_concern Estimate Std. Error t value Pr(>|t|) (Intercept) 86.79180 0.778064 111.5485 < 2.2e-16 *** conditiontreatment 2.18555 1.117468 1.9558 0.050723 . --- Dep. var.: conspiracy Estimate Std. Error t value Pr(>|t|) (Intercept) 78.13396 0.535981 145.77751 < 2.2e-16 *** conditiontreatment 1.15773 0.769785 1.50397 0.13286 --- Dep. var.: aot Estimate Std. Error t value Pr(>|t|) (Intercept) 3.323060 0.045072 73.727253 < 2.2e-16 *** conditiontreatment 0.051911 0.064734 0.801915 0.42276 ``` causal forests doubly robust estimates ```r # india b = -0.15 [-0.22, -0.08], p < .001 # philippines b = -0.16 [-0.22, -0.10], p < .001 ``` Summary For the india/philippines experiments on lucid, I'm going to briefly report something similar. Attrition at each stage didn't differ by condition (chi-square tests at all stages of the survey are ps > .175): - those who passed vs failed initial bot checks (ps for both countries >= .520) - those who passed the first attention check shown immediately after consent (ps >= .463) - those who passed the second attention check shown with the questionnaires/surveys (ps >= .175) - those who completed the exp (ps > .200) Then tested whether each covariate differed by condition in each country. Only in India did we find differences, and again they're measured post-treatment so could be driven by post-treatment effects: - belief in god (lower in treatment), p = .021 - AOT (higher in treatment), p = .006 And finally, report causal forests robust ATEs: - India: b = -0.15 [-0.22, -0.08] - philippines: b = -0.16 [-0.22, -0.10]