IAT regressions with Age
Molly Lewis
2018-06-12
These analyses predict iat behavioral, language iat measure, objective equality measures (ggi, wps, percentage women in stem). GGI and WPS are standard measures of gender equality with larger numbers indicating higher equality. Percentage women in stem comes from Stoet and Geary paper (there are also misc other objective variables from this paper)
The analyses are done at three different levels - participant, country, and language.
The resid values are the iat scores (impicit and explicit) residualizing out age, sex (resid1), and age, sex relgiosity, and explicit/implicit score (resid2).
Participant level findings
- When you control for age of participants, ggi predicts iat scores in the same direciton we found before: more equal countryies have higher behavioral iat. The same pattern holds for wps.
- Language iat also predicts behavioral iat, controling for ggi/wps and explicit scores
Country level findings
- GGI predicts behavioral iat, controling for stuff in regression, in the predicted direction: higher ggi, lower bias. But effect of ggi goes away whne you look at residuals.
- However, for WPS, the effect is in the same direction as at the participant level: higher wps, higher behavioral bias.
- GGI predicts women in stem measure in same direction as participant level: higher ggi, lower percentage of women in stem (this is the Stoet and Geary 2018 effect)
- Importantly, es_behavioral_iat also predict percentage of women in stem, controling for ggi and explicit beliefs: higher iat bias, lower percentage of women in stem.
So the relationship between objective measures of equality and behavioral iat is the same between participant and country level (more equality, more bias), except for one regression (iat~ggi + age etc.).
Language level findings
- career language iat does not predict behavioral iat, when you control for mean age of participants, but the effect is in the predicted direction
- flower language iat does not predict behavioral iat at all
- career language iat does not predict percentage women measure, but does predict satisfaction (from Stoet and Geary), even after controlling for ggi
So, in sum:
- At the participant level, language iat predicts behavioral iat controling for explicit bias and objective measures (more language iat bias, more behavioral iat bias)
- As in the cogsci paper, objective equality is positively associated with iat bias at the participant level (more equal, more bias). This is true from multiple measures (GGI and WPS).
- With minor complications, we see the same relationship as at the country level between objective equality and iat behavioral bias (more equal, more bias).
- Replicating Stoet and Geary, we also find at the country level that GGI is negatively associated with percentage women in stem (more equal, less stem) - but there’s an independent contribution of behavior iat in the predicted direction (more iat bias, less stem).
- Nothing is significant at the language level (though might try weighted averages here, as before).
PARTICIPANT_DF_PATH <- "by_participant_df.csv"
COUNTRY_DF_PATH <- "by_country_df.csv"
LANGUAGE_DF_PATH <- "by_language_df.csv"
participant_df_raw <- read_csv(PARTICIPANT_DF_PATH)
participant_df <- participant_df_raw %>%
mutate(es_iat_sex_age_religion_explicit_resid = as.numeric(es_iat_sex_age_religion_explicit_resid),
es_explicit_sex_age_religion_iat_resid = as.numeric( es_explicit_sex_age_religion_iat_resid)) %>%
mutate_at(vars(education, religionid,), as.numeric) %>%
mutate(age_log = log(age)) %>%
mutate_at(vars(overall_iat_D_score, age, religionid, explicit_dif, es_iat_sex_age_resid, es_iat_sex_age_religion_explicit_resid, es_explicit_sex_age_implicit_resid,
es_explicit_sex_age_religion_iat_resid), scale_this)
country_df <- read_csv(COUNTRY_DF_PATH) %>%
mutate(es_behavioral_iat_resid2 = as.numeric(es_behavioral_iat_resid2),
es_behavioral_iat_resid1 = as.numeric(es_behavioral_iat_resid1)) %>%
mutate_if(is.numeric, scale_this) %>%
select(-ggi_stoet)
language_df <- read_csv(LANGUAGE_DF_PATH) %>%
mutate_if(is.numeric, scale_this) %>%
select(-weapons_google, -career_google)Participant Level
participant_df_full <- participant_df %>%
left_join(country_df %>% select(country_code:satisfaction, wiki_language_code),
by = c("countryres" = "country_code")) %>%
left_join(language_df %>% select(career_hand,flowers_google, wiki_language_code))Raw correlations
participant_df_corr <- participant_df_full %>%
select_if(is.numeric)
corr_mat <- cor(participant_df_corr,
use = "pairwise.complete.obs")
p.mat <- corrplot::cor.mtest(participant_df_corr,
conf.level = (1-ALPHA),
use = "pairwise.complete.obs")$p
corrplot::corrplot(corr_mat, method = "color", col = cols,
order = "original", number.cex = .7,
addCoef.col = "black",
p.mat = p.mat, sig.level = ALPHA, insig = "blank",
tl.col = "black", tl.srt = 90,
diag = FALSE)
Regressions
ggi predicts iat score
lm(overall_iat_D_score ~ ggi + age_log,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = overall_iat_D_score ~ ggi + age_log, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.3271 -0.6486 0.0626 0.7068 3.9229
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.521775 0.011361 -45.925 <2e-16 ***
## ggi 0.004955 0.004667 1.062 0.288
## age_log 0.159326 0.003419 46.600 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9973 on 633964 degrees of freedom
## (29742 observations deleted due to missingness)
## Multiple R-squared: 0.003427, Adjusted R-squared: 0.003424
## F-statistic: 1090 on 2 and 633964 DF, p-value: < 2.2e-16lm(overall_iat_D_score ~ ggi + age_log + religionid + as.factor(sex) ,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = overall_iat_D_score ~ ggi + age_log + religionid +
## as.factor(sex), data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.3077 -0.6384 0.0572 0.6909 3.5071
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.462288 0.013462 -34.34 < 2e-16 ***
## ggi 0.030926 0.006567 4.71 2.48e-06 ***
## age_log 0.173352 0.004022 43.10 < 2e-16 ***
## religionid 0.026226 0.001504 17.44 < 2e-16 ***
## as.factor(sex)1 -0.332222 0.003384 -98.16 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9807 on 428499 degrees of freedom
## (235205 observations deleted due to missingness)
## Multiple R-squared: 0.02601, Adjusted R-squared: 0.02601
## F-statistic: 2861 on 4 and 428499 DF, p-value: < 2.2e-16lm(overall_iat_D_score ~ ggi + age_log + religionid + as.factor(sex) + explicit_dif ,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = overall_iat_D_score ~ ggi + age_log + religionid +
## as.factor(sex) + explicit_dif, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.3404 -0.6298 0.0557 0.6816 3.5516
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.465285 0.013420 -34.671 < 2e-16 ***
## ggi 0.026584 0.006552 4.057 4.96e-05 ***
## age_log 0.173082 0.004007 43.190 < 2e-16 ***
## religionid 0.018239 0.001499 12.166 < 2e-16 ***
## as.factor(sex)1 -0.334838 0.003371 -99.341 < 2e-16 ***
## explicit_dif 0.138112 0.001419 97.350 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9702 on 423235 degrees of freedom
## (240468 observations deleted due to missingness)
## Multiple R-squared: 0.04709, Adjusted R-squared: 0.04708
## F-statistic: 4183 on 5 and 423235 DF, p-value: < 2.2e-16lm(es_iat_sex_age_resid ~ ggi ,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = es_iat_sex_age_resid ~ ggi, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.4848 -0.6494 0.0598 0.7043 4.1454
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.005266 0.002689 -1.959 0.0502 .
## ggi 0.009625 0.004671 2.061 0.0393 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9988 on 633965 degrees of freedom
## (29742 observations deleted due to missingness)
## Multiple R-squared: 6.699e-06, Adjusted R-squared: 5.122e-06
## F-statistic: 4.247 on 1 and 633965 DF, p-value: 0.03932lm(es_iat_sex_age_resid ~ wps ,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = es_iat_sex_age_resid ~ wps, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.4853 -0.6493 0.0598 0.7045 4.1449
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.042230 0.004910 -8.601 <2e-16 ***
## wps 0.035758 0.004063 8.802 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9988 on 634828 degrees of freedom
## (28879 observations deleted due to missingness)
## Multiple R-squared: 0.000122, Adjusted R-squared: 0.0001204
## F-statistic: 77.47 on 1 and 634828 DF, p-value: < 2.2e-16language also iat predicts iat score
lm(overall_iat_D_score ~ career_hand + ggi + age_log,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = overall_iat_D_score ~ career_hand + ggi + age_log,
## data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.3305 -0.6482 0.0629 0.7066 3.9197
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.527970 0.011999 -44.001 < 2e-16 ***
## career_hand 0.023710 0.005934 3.996 6.44e-05 ***
## ggi -0.010870 0.005707 -1.905 0.0568 .
## age_log 0.159542 0.003453 46.202 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9971 on 617474 degrees of freedom
## (46231 observations deleted due to missingness)
## Multiple R-squared: 0.00347, Adjusted R-squared: 0.003465
## F-statistic: 716.7 on 3 and 617474 DF, p-value: < 2.2e-16lm(es_iat_sex_age_resid ~ career_hand + ggi ,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = es_iat_sex_age_resid ~ career_hand + ggi, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.4866 -0.6490 0.0599 0.7040 4.1435
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.012557 0.004498 -2.791 0.00525 **
## career_hand 0.016569 0.005943 2.788 0.00530 **
## ggi 0.004923 0.005710 0.862 0.38862
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9987 on 617475 degrees of freedom
## (46231 observations deleted due to missingness)
## Multiple R-squared: 1.714e-05, Adjusted R-squared: 1.39e-05
## F-statistic: 5.291 on 2 and 617475 DF, p-value: 0.005039lm(es_iat_sex_age_resid ~ career_hand + ggi + es_explicit_sex_age_implicit_resid,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = es_iat_sex_age_resid ~ career_hand + ggi + es_explicit_sex_age_implicit_resid,
## data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.4371 -0.6406 0.0575 0.6949 4.1966
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.007586 0.004491 -1.689 0.0912
## career_hand 0.012163 0.005926 2.052 0.0401
## ggi 0.003260 0.005688 0.573 0.5665
## es_explicit_sex_age_implicit_resid 0.157157 0.001265 124.231 <2e-16
##
## (Intercept) .
## career_hand *
## ggi
## es_explicit_sex_age_implicit_resid ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9865 on 605858 degrees of freedom
## (57847 observations deleted due to missingness)
## Multiple R-squared: 0.02486, Adjusted R-squared: 0.02485
## F-statistic: 5148 on 3 and 605858 DF, p-value: < 2.2e-16lm(es_iat_sex_age_resid ~ career_hand + wps,
data = participant_df_full) %>%
summary()##
## Call:
## lm(formula = es_iat_sex_age_resid ~ career_hand + wps, data = participant_df_full)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.487 -0.649 0.060 0.704 4.144
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.023902 0.007247 -3.298 0.000972 ***
## career_hand 0.015022 0.005998 2.505 0.012262 *
## wps 0.012456 0.006133 2.031 0.042252 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9987 on 618338 degrees of freedom
## (45368 observations deleted due to missingness)
## Multiple R-squared: 2.472e-05, Adjusted R-squared: 2.149e-05
## F-statistic: 7.644 on 2 and 618338 DF, p-value: 0.0004791Country Level
Raw correlations
country_df_corr <- country_df %>%
select_if(is.numeric)
corr_mat <- cor(country_df_corr,
use = "pairwise.complete.obs")
p.mat <- corrplot::cor.mtest(country_df_corr,
conf.level = (1-ALPHA),
use = "pairwise.complete.obs")$p
corrplot::corrplot(corr_mat, method = "color", col = cols,
order = "original", number.cex = .7,
addCoef.col = "black",
p.mat = p.mat, sig.level = ALPHA, insig = "blank",
tl.col = "black", tl.srt = 90,
diag = FALSE)
Regressions
Predicting IAT Scores
Controling for stuff, objective measure ggi is correlated with iat in the predicted direction: more objectively equal, less bias. Wps is predictive in the same direction when you control for age etc. by residualizing it out.
ggi
lm(es_behavioral_iat ~ ggi + participant_sex + participant_age + participant_religosity + es_behavioral_explicit,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ ggi + participant_sex + participant_age +
## participant_religosity + es_behavioral_explicit, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.32950 -0.47106 -0.02099 0.40092 1.51753
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1970 0.1309 1.505 0.140742
## ggi -0.3836 0.1491 -2.572 0.014255 *
## participant_sex -0.4452 0.1311 -3.396 0.001647 **
## participant_age 0.3756 0.1645 2.283 0.028299 *
## participant_religosity -0.3004 0.1272 -2.361 0.023616 *
## es_behavioral_explicit 0.4954 0.1370 3.616 0.000886 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7146 on 37 degrees of freedom
## (193 observations deleted due to missingness)
## Multiple R-squared: 0.5673, Adjusted R-squared: 0.5088
## F-statistic: 9.701 on 5 and 37 DF, p-value: 5.567e-06lm(es_behavioral_iat ~ ggi + es_behavioral_explicit ,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ ggi + es_behavioral_explicit,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.34493 -0.36726 -0.04691 0.61136 1.54413
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.07831 0.14774 0.530 0.599
## ggi -0.16843 0.15531 -1.084 0.285
## es_behavioral_explicit 0.65483 0.13760 4.759 2.54e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8329 on 40 degrees of freedom
## (193 observations deleted due to missingness)
## Multiple R-squared: 0.3645, Adjusted R-squared: 0.3327
## F-statistic: 11.47 on 2 and 40 DF, p-value: 0.0001155ggi-residuals
lm(es_behavioral_iat_resid1 ~ ggi + es_behavioral_explicit_resid1,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ ggi + es_behavioral_explicit_resid1,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.08866 -0.46497 0.04488 0.55194 2.05616
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.02288 0.17522 -0.131 0.8967
## ggi 0.09358 0.17623 0.531 0.5983
## es_behavioral_explicit_resid1 0.34017 0.15029 2.263 0.0291 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9876 on 40 degrees of freedom
## (193 observations deleted due to missingness)
## Multiple R-squared: 0.1139, Adjusted R-squared: 0.0696
## F-statistic: 2.571 on 2 and 40 DF, p-value: 0.08905lm(es_behavioral_iat_resid2 ~ ggi + es_behavioral_explicit_resid2,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid2 ~ ggi + es_behavioral_explicit_resid2,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.92618 -0.53342 -0.02261 0.65292 2.40690
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.04014 0.18257 -0.220 0.827
## ggi 0.12988 0.18191 0.714 0.479
## es_behavioral_explicit_resid2 0.14065 0.16403 0.857 0.396
##
## Residual standard error: 1.034 on 40 degrees of freedom
## (193 observations deleted due to missingness)
## Multiple R-squared: 0.03162, Adjusted R-squared: -0.0168
## F-statistic: 0.6531 on 2 and 40 DF, p-value: 0.5259lm(es_behavioral_iat_resid1 ~ ggi ,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ ggi, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.00089 -0.71342 -0.07046 0.77283 2.35158
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.02204 0.18263 0.121 0.905
## ggi 0.02373 0.18202 0.130 0.897
##
## Residual standard error: 1.036 on 41 degrees of freedom
## (193 observations deleted due to missingness)
## Multiple R-squared: 0.0004144, Adjusted R-squared: -0.02397
## F-statistic: 0.017 on 1 and 41 DF, p-value: 0.8969wps
lm(es_behavioral_iat ~ wps + participant_sex + participant_age + participant_religosity + es_behavioral_explicit,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ wps + participant_sex + participant_age +
## participant_religosity + es_behavioral_explicit, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.33347 -0.53731 -0.03152 0.55331 1.69796
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.05305 0.16187 -0.328 0.74494
## wps 0.08145 0.16028 0.508 0.61427
## participant_sex -0.32031 0.13715 -2.336 0.02490 *
## participant_age 0.16766 0.16650 1.007 0.32033
## participant_religosity -0.25887 0.13764 -1.881 0.06768 .
## es_behavioral_explicit 0.43123 0.15228 2.832 0.00736 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7649 on 38 degrees of freedom
## (192 observations deleted due to missingness)
## Multiple R-squared: 0.4969, Adjusted R-squared: 0.4307
## F-statistic: 7.506 on 5 and 38 DF, p-value: 5.554e-05wps-residuals
lm(es_behavioral_iat_resid1 ~ wps + es_behavioral_explicit_resid1,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ wps + es_behavioral_explicit_resid1,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7466 -0.5119 -0.0196 0.5634 2.0836
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.3138 0.1725 -1.819 0.07625 .
## wps 0.4454 0.1489 2.992 0.00468 **
## es_behavioral_explicit_resid1 0.3373 0.1336 2.524 0.01556 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8919 on 41 degrees of freedom
## (192 observations deleted due to missingness)
## Multiple R-squared: 0.2651, Adjusted R-squared: 0.2293
## F-statistic: 7.395 on 2 and 41 DF, p-value: 0.001809lm(es_behavioral_iat_resid1 ~ wps ,
data = country_df) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ wps, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.6239 -0.5687 -0.1640 0.7153 2.4369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.2923 0.1830 -1.598 0.11764
## wps 0.4317 0.1580 2.732 0.00917 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9472 on 42 degrees of freedom
## (192 observations deleted due to missingness)
## Multiple R-squared: 0.1509, Adjusted R-squared: 0.1307
## F-statistic: 7.464 on 1 and 42 DF, p-value: 0.009166Predicting women in stem
Both iat and ggi are predictors of percentage women in stem. The ggi measure is in the same direction as Stoet. The IAT measure is in the predicted direction: more biased, few women in stem. Explicit judgement do not predict percentage of women in stem. It does not predict any of the other measures for Stoet paper.
model with controls
lm(per_women_stem ~ es_behavioral_iat + ggi+ es_behavioral_explicit + participant_sex + participant_age + participant_religosity ,
data = country_df) %>%
summary()##
## Call:
## lm(formula = per_women_stem ~ es_behavioral_iat + ggi + es_behavioral_explicit +
## participant_sex + participant_age + participant_religosity,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.01740 -0.38235 -0.05711 0.32865 1.20424
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.33251 0.15192 2.189 0.03859 *
## es_behavioral_iat -0.38704 0.17343 -2.232 0.03524 *
## ggi -0.53733 0.17205 -3.123 0.00462 **
## es_behavioral_explicit 0.14002 0.16647 0.841 0.40859
## participant_sex -0.20763 0.16400 -1.266 0.21765
## participant_age 0.08367 0.17793 0.470 0.64243
## participant_religosity 0.02435 0.19472 0.125 0.90153
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6282 on 24 degrees of freedom
## (205 observations deleted due to missingness)
## Multiple R-squared: 0.5129, Adjusted R-squared: 0.3912
## F-statistic: 4.212 on 6 and 24 DF, p-value: 0.004933residuals
lm(per_women_stem ~ es_behavioral_iat_resid1 + ggi + es_behavioral_explicit_resid1,
data = country_df) %>%
summary()##
## Call:
## lm(formula = per_women_stem ~ es_behavioral_iat_resid1 + ggi +
## es_behavioral_explicit_resid1, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.90882 -0.28145 -0.02449 0.39670 1.13701
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.27505 0.11442 2.404 0.023354 *
## es_behavioral_iat_resid1 -0.44082 0.10480 -4.206 0.000256 ***
## ggi -0.43199 0.10709 -4.034 0.000404 ***
## es_behavioral_explicit_resid1 0.15214 0.09734 1.563 0.129684
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5327 on 27 degrees of freedom
## (205 observations deleted due to missingness)
## Multiple R-squared: 0.606, Adjusted R-squared: 0.5622
## F-statistic: 13.84 on 3 and 27 DF, p-value: 1.174e-05lm(per_women_stem ~ es_behavioral_iat_resid1 + ggi,
data = country_df) %>%
summary()##
## Call:
## lm(formula = per_women_stem ~ es_behavioral_iat_resid1 + ggi,
## data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.91065 -0.33365 -0.05943 0.31671 1.20908
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.28291 0.11722 2.414 0.022590 *
## es_behavioral_iat_resid1 -0.37451 0.09827 -3.811 0.000696 ***
## ggi -0.46821 0.10721 -4.367 0.000156 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5462 on 28 degrees of freedom
## (205 observations deleted due to missingness)
## Multiple R-squared: 0.5703, Adjusted R-squared: 0.5396
## F-statistic: 18.58 on 2 and 28 DF, p-value: 7.314e-06lm(per_women_stem ~ es_behavioral_iat_resid1 + wps + es_behavioral_explicit_resid1,
data = country_df) %>%
summary()##
## Call:
## lm(formula = per_women_stem ~ es_behavioral_iat_resid1 + wps +
## es_behavioral_explicit_resid1, data = country_df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.18898 -0.38351 -0.08598 0.29577 1.12183
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.3993 0.2020 1.977 0.05833 .
## es_behavioral_iat_resid1 -0.3952 0.1317 -3.001 0.00573 **
## wps -0.4039 0.1824 -2.215 0.03540 *
## es_behavioral_explicit_resid1 0.2099 0.1114 1.885 0.07020 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6204 on 27 degrees of freedom
## (205 observations deleted due to missingness)
## Multiple R-squared: 0.4656, Adjusted R-squared: 0.4062
## F-statistic: 7.84 on 3 and 27 DF, p-value: 0.00064Language Level
Raw correlations
language_df_with_ggi <- country_df %>%
select(ggi, wps, wiki_language_code,es_behavioral_iat_resid2, es_behavioral_iat_resid1) %>%
group_by(wiki_language_code) %>%
summarize_all(mean, na.rm = T) %>%
left_join(language_df %>% select(-es_behavioral_iat_resid2, -es_behavioral_iat_resid1)) %>%
select(-contains("weighted"), -contains("stoet"))
language_df_corr <- language_df_with_ggi %>%
select_if(is.numeric)
corr_mat <- cor(language_df_corr,
use = "pairwise.complete.obs")
p.mat <- corrplot::cor.mtest(language_df_corr,
conf.level = (1-ALPHA),
use = "pairwise.complete.obs")$p
corrplot::corrplot(corr_mat, method = "color", col = cols,
order = "original", number.cex = .7,
addCoef.col = "black",
p.mat = p.mat, sig.level = ALPHA, insig = "blank",
tl.col = "black", tl.srt = 90,
diag = FALSE)
Regressions predicting behavioral iat
Career language
behvioral iat is correlated with language measure, but goes away when you control for mean age
model with controls
lm(es_behavioral_iat ~ career_hand,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ career_hand, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.21066 -0.58248 0.09664 0.43002 1.45541
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1356 0.1765 0.769 0.453
## career_hand 0.3660 0.1816 2.015 0.061 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7488 on 16 degrees of freedom
## (14 observations deleted due to missingness)
## Multiple R-squared: 0.2025, Adjusted R-squared: 0.1526
## F-statistic: 4.062 on 1 and 16 DF, p-value: 0.06098lm(es_behavioral_iat ~ career_hand + participant_age ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ career_hand + participant_age,
## data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.28063 -0.33293 0.03798 0.25808 1.19697
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.001133 0.151832 0.007 0.99414
## career_hand 0.028335 0.187758 0.151 0.88205
## participant_age 0.640442 0.216536 2.958 0.00978 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6146 on 15 degrees of freedom
## (14 observations deleted due to missingness)
## Multiple R-squared: 0.4963, Adjusted R-squared: 0.4291
## F-statistic: 7.388 on 2 and 15 DF, p-value: 0.005842lm(es_behavioral_iat ~ career_hand + ggi + participant_age + participant_sex + + participant_religosity + es_behavioral_explicit,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat ~ career_hand + ggi + participant_age +
## participant_sex + +participant_religosity + es_behavioral_explicit,
## data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.52923 -0.20804 -0.05642 0.21805 0.69073
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1381 0.1426 0.968 0.3558
## career_hand 0.1746 0.1601 1.090 0.3012
## ggi -0.3974 0.1590 -2.499 0.0315 *
## participant_age 0.4779 0.2294 2.083 0.0639 .
## participant_sex -0.4300 0.2373 -1.812 0.1001
## participant_religosity -0.3725 0.1770 -2.105 0.0616 .
## es_behavioral_explicit 0.3407 0.1655 2.058 0.0666 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4795 on 10 degrees of freedom
## (15 observations deleted due to missingness)
## Multiple R-squared: 0.7851, Adjusted R-squared: 0.6562
## F-statistic: 6.09 on 6 and 10 DF, p-value: 0.006489Residuals
lm(es_behavioral_iat_resid1 ~ career_hand,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ career_hand, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.75133 -0.59491 0.01759 0.57257 1.29088
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.05597 0.20388 0.275 0.787
## career_hand 0.32889 0.20979 1.568 0.137
##
## Residual standard error: 0.865 on 16 degrees of freedom
## (14 observations deleted due to missingness)
## Multiple R-squared: 0.1332, Adjusted R-squared: 0.07898
## F-statistic: 2.458 on 1 and 16 DF, p-value: 0.1365lm(es_behavioral_iat_resid2 ~ career_hand,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid2 ~ career_hand, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7416 -0.5723 0.1985 0.6177 1.1716
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.01993 0.21391 0.093 0.927
## career_hand 0.23343 0.22011 1.061 0.305
##
## Residual standard error: 0.9075 on 16 degrees of freedom
## (14 observations deleted due to missingness)
## Multiple R-squared: 0.06568, Adjusted R-squared: 0.007283
## F-statistic: 1.125 on 1 and 16 DF, p-value: 0.3047lm(es_behavioral_iat_resid1 ~ career_hand + es_behavioral_explicit_resid1 + ggi ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ career_hand + es_behavioral_explicit_resid1 +
## ggi, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.84998 -0.28282 0.05293 0.52773 1.38202
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.06749 0.24689 -0.273 0.789
## career_hand 0.16006 0.24562 0.652 0.526
## es_behavioral_explicit_resid1 0.28702 0.24106 1.191 0.255
## ggi 0.28268 0.24786 1.140 0.275
##
## Residual standard error: 0.8795 on 13 degrees of freedom
## (15 observations deleted due to missingness)
## Multiple R-squared: 0.2495, Adjusted R-squared: 0.07635
## F-statistic: 1.441 on 3 and 13 DF, p-value: 0.2761lm(es_behavioral_iat_resid2 ~ career_hand + ggi ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid2 ~ career_hand + ggi, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.6840 -0.4930 0.2261 0.5690 1.3079
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.08812 0.23557 -0.374 0.714
## career_hand 0.09178 0.22948 0.400 0.695
## ggi 0.34858 0.22855 1.525 0.149
##
## Residual standard error: 0.8707 on 14 degrees of freedom
## (15 observations deleted due to missingness)
## Multiple R-squared: 0.1999, Adjusted R-squared: 0.08556
## F-statistic: 1.749 on 2 and 14 DF, p-value: 0.21Flowers language
Flowers not predictive.
lm(es_behavioral_iat_resid1 ~ flowers_google + es_behavioral_explicit_resid1 + ggi ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = es_behavioral_iat_resid1 ~ flowers_google + es_behavioral_explicit_resid1 +
## ggi, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.18442 -0.52696 0.03634 0.49360 1.78979
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.07444 0.21709 0.343 0.7354
## flowers_google 0.20774 0.25041 0.830 0.4171
## es_behavioral_explicit_resid1 0.37039 0.20495 1.807 0.0866 .
## ggi 0.14214 0.21813 0.652 0.5225
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9658 on 19 degrees of freedom
## (9 observations deleted due to missingness)
## Multiple R-squared: 0.1597, Adjusted R-squared: 0.02699
## F-statistic: 1.203 on 3 and 19 DF, p-value: 0.3354Regressions Predicting Stoet measures
women in stem
lm(per_women_stem ~ ggi+ career_hand ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = per_women_stem ~ ggi + career_hand, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1753 -0.2426 0.0417 0.5181 0.7913
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.419735 0.198945 2.110 0.05860 .
## ggi -0.725624 0.177116 -4.097 0.00177 **
## career_hand 0.002792 0.184412 0.015 0.98819
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6626 on 11 degrees of freedom
## (18 observations deleted due to missingness)
## Multiple R-squared: 0.6264, Adjusted R-squared: 0.5585
## F-statistic: 9.223 on 2 and 11 DF, p-value: 0.004446satisfaction
lm(satisfaction ~ ggi + career_hand ,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = satisfaction ~ ggi + career_hand, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.0251 -0.2948 0.1064 0.3189 0.7068
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.06957 0.13531 -0.514 0.61519
## ggi 0.48354 0.13128 3.683 0.00246 **
## career_hand 0.41321 0.13181 3.135 0.00731 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5001 on 14 degrees of freedom
## (15 observations deleted due to missingness)
## Multiple R-squared: 0.7321, Adjusted R-squared: 0.6939
## F-statistic: 19.13 on 2 and 14 DF, p-value: 9.895e-05lm(satisfaction ~ ggi+ flowers_google,
data = language_df_with_ggi) %>%
summary()##
## Call:
## lm(formula = satisfaction ~ ggi + flowers_google, data = language_df_with_ggi)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.09834 -0.39545 -0.08244 0.36479 1.17247
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.30919 0.14989 -2.063 0.0531 .
## ggi 0.75332 0.14436 5.218 4.9e-05 ***
## flowers_google -0.09499 0.16304 -0.583 0.5670
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6456 on 19 degrees of freedom
## (10 observations deleted due to missingness)
## Multiple R-squared: 0.6072, Adjusted R-squared: 0.5658
## F-statistic: 14.69 on 2 and 19 DF, p-value: 0.0001395