Key Findings Summary

White Population:

Internet Users:

The temporal volatility of hate speech is not significantly associated with any of the mental health outcomes.

For non-internet users:

Increased variability in hate speech is associated with fewer bad mental health days (6-day decrease) and a lower likelihood of experiencing any or frequent bad mental health days.

Variability in VADER scores shows no significant association with mental health outcomes.

Latinx Population:

For internet users:

Higher variability in hate speech is associated with more bad mental health days (10.3-day increase) but not with “any” or “frequent” bad mental health days. Variability in VADER scores shows no significant association with mental health outcomes.

For non-internet users:

There is no significant relationship between variability in hate speech or VADER scores and any of the mental health outcomes.

Poisson Regression Results (Latinx Population):

For mental health days, a one-unit increase in the variability of hate speech over 90 days is associated with an 18.04-fold increase in expected poor mental health days. For “any” or “frequent” bad mental health days, significant results are suggested, but interpretation is hindered due to model convergence issues.

Descriptives

Mean Mental Health Days by Race (Unweighted)
Race_Label Mean_Poor_Menthlth_Days Mean_Any_Bad_Menthlth_Days Mean_Frequent_Bad_Menthlth_Days
Latino 3.829876 0.3219597 0.1226691
White 3.381298 0.3147087 0.1042163
Mean Mental Health Days by Race (Weighted)
Race_Label Mean_Poor_Menthlth_Days Mean_Any_Bad_Menthlth_Days Mean_Frequent_Bad_Menthlth_Days
Latino 3.694928 0.3301296 0.1166011
White 3.931572 0.3586166 0.1221424

Internet

White Models

Number of Days

Regression results examining rolling average standard deviation and number of bad mental health days - White
term estimate std.error statistic p.value
(Intercept) 6.499 1.654 3.929 0.000
X_age_g -0.388 0.005 -75.743 0.000
X_educag -0.889 0.008 -109.259 0.000
female 1.557 0.015 103.186 0.000
INCOME2 -0.009 0.000 -32.385 0.000
MARITAL 0.354 0.005 75.808 0.000
rolling_sd_HS90 -0.011 1.305 -0.008 0.993

The recent temporal volatility of hate speech is not associated to white mental health

Any Bad Days

Regression results examining rolling average standard deviation and any bad mental health - White
term estimate std.error statistic p.value
(Intercept) 0.468 0.100 4.686 0.000
X_age_g -0.050 0.000 -160.687 0.000
X_educag -0.015 0.000 -31.302 0.000
female 0.130 0.001 142.477 0.000
INCOME2 -0.001 0.000 -46.808 0.000
MARITAL 0.023 0.000 80.997 0.000
rolling_sd_HS90 0.002 0.079 0.022 0.983

The recent temporal volatility of hate speech is not associated to white mental health for any bad days mental health

Frequent Bad Mental Health

Regression results examining rolling average standard deviation and frequent bad mental health - White
term estimate std.error statistic p.value
(Intercept) 0.239 0.067 3.540 0.000
X_age_g -0.011 0.000 -51.945 0.000
X_educag -0.035 0.000 -106.625 0.000
female 0.051 0.001 82.768 0.000
INCOME2 0.000 0.000 -24.582 0.000
MARITAL 0.012 0.000 63.192 0.000
rolling_sd_HS90 -0.020 0.053 -0.369 0.712

The recent temporal volatility of hate speech is not associated to white mental health for frequent bad days mental health

Latinx Model

Number of Days

Regression results examining rolling average standard deviation and mental health - Latino
term estimate std.error statistic p.value
(Intercept) -9.650 5.934 -1.626 0.104
X_age_g -0.080 0.021 -3.745 0.000
X_educag -0.060 0.024 -2.458 0.014
female 1.033 0.052 19.786 0.000
INCOME2 -0.008 0.001 -7.882 0.000
MARITAL 0.289 0.014 20.112 0.000
rolling_sd_HS90 10.271 4.417 2.325 0.020

The recent temporal volatility of hate speech is associated with a 10.3 day increase in bad mental health days for the Latinx population.

Any Bad Days

Regression results examining rolling average standard deviation and any bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 0.203 0.373 0.546 0.585
X_age_g -0.028 0.001 -21.033 0.000
X_educag 0.013 0.002 8.423 0.000
female 0.075 0.003 22.858 0.000
INCOME2 -0.001 0.000 -12.703 0.000
MARITAL 0.021 0.001 23.488 0.000
rolling_sd_HS90 0.124 0.277 0.448 0.654

The recent temporal volatility of hate speech is not associated to any bad days mental health for the Latinx population.

Frequent Bad Days

Regression results examining rolling average standard deviation and Frequent bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) -0.354 0.245 -1.443 0.149
X_age_g -0.001 0.001 -1.446 0.148
X_educag -0.004 0.001 -4.041 0.000
female 0.035 0.002 16.129 0.000
INCOME2 0.000 0.000 -5.978 0.000
MARITAL 0.009 0.001 14.985 0.000
rolling_sd_HS90 0.341 0.183 1.867 0.062

The recent temporal volatility of hate speech is not associated to frequent bad days mental health for the Latinx population.

No Internet

White Models

Number of Days

Regression results examining rolling average standard deviation and days of bad mental health - White
term estimate std.error statistic p.value
(Intercept) 19.853 4.291 4.627 0.000
X_age_g -1.102 0.017 -65.592 0.000
X_educag -0.618 0.020 -30.992 0.000
female 1.153 0.039 29.898 0.000
INCOME2 -0.013 0.001 -23.366 0.000
MARITAL 0.212 0.013 15.775 0.000
rolling_sd_HS90 -6.858 3.395 -2.020 0.043

The recent temporal volatility of hate speech is associated to a 6 day decrease in bad days mental health for the white population who does not use the internet.

Any Bad Days

Regression results examining rolling average standard deviation and any days bad mental health - White
term estimate std.error statistic p.value
(Intercept) 1.064 0.216 4.932 0.000
X_age_g -0.057 0.001 -67.488 0.000
X_educag -0.020 0.001 -19.827 0.000
female 0.087 0.002 44.646 0.000
INCOME2 -0.001 0.000 -30.887 0.000
MARITAL 0.011 0.001 15.898 0.000
rolling_sd_HS90 -0.361 0.171 -2.115 0.034

The recent temporal volatility of hate speech is associated to a 36% day decrease in any bad days mental health for the white population who does not use the internet.

Frequent Bad Days

Regression results examining rolling average standard deviation and frequent bad mental health - White
term estimate std.error statistic p.value
(Intercept) 0.738 0.166 4.450 0.000
X_age_g -0.039 0.001 -60.670 0.000
X_educag -0.024 0.001 -30.557 0.000
female 0.038 0.001 25.687 0.000
INCOME2 0.000 0.000 -20.613 0.000
MARITAL 0.007 0.001 13.916 0.000
rolling_sd_HS90 -0.267 0.131 -2.033 0.042

The recent temporal volatility of hate speech is associated to a 27% day decrease in any bad days mental health for the white population who does not use the internet.

Latinx Model

Number of Days

Regression results examining rolling average standard deviation and days of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 5.349 10.716 0.499 0.618
X_age_g 0.369 0.034 10.846 0.000
X_educag -0.143 0.053 -2.681 0.007
female 0.414 0.095 4.367 0.000
INCOME2 -0.002 0.002 -0.967 0.333
MARITAL 0.124 0.026 4.716 0.000
rolling_sd_HS90 -0.964 8.172 -0.118 0.906

No association between The recent temporal volatility of hate speech and number of days bad mental health for Latinos that don’t use the internet.

Any Bad Days

Regression results examining rolling average standard deviation and any days of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 0.053 0.585 0.091 0.928
X_age_g 0.004 0.002 2.423 0.015
X_educag -0.009 0.003 -2.967 0.003
female 0.033 0.005 6.338 0.000
INCOME2 0.000 0.000 -5.253 0.000
MARITAL 0.011 0.001 7.987 0.000
rolling_sd_HS90 0.121 0.446 0.272 0.786

No association between The recent temporal volatility of hate speech and any days bad mental health for Latinos that don’t use the internet.

Frequent Bad Days

Regression results examining rolling average standard deviation and frequent days of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 0.015 0.427 0.034 0.973
X_age_g 0.014 0.001 9.993 0.000
X_educag -0.004 0.002 -1.862 0.063
female 0.008 0.004 2.247 0.025
INCOME2 0.000 0.000 -0.902 0.367
MARITAL 0.004 0.001 3.641 0.000
rolling_sd_HS90 0.089 0.325 0.273 0.785

No association between The recent temporal volatility of hate speech and frequent days bad mental health for Latinos that don’t use the internet.

VADER

White Models

Number of Days

Regression results examining rolling average standard deviation and days of bad mental health - White
term estimate std.error statistic p.value
(Intercept) 6.059 2.490 2.434 0.015
X_age_g -0.388 0.005 -75.739 0.000
X_educag -0.889 0.008 -109.259 0.000
female 1.557 0.015 103.186 0.000
INCOME2 -0.009 0.000 -32.385 0.000
MARITAL 0.354 0.005 75.808 0.000
rolling_sd_VADER90 1.119 6.507 0.172 0.863

Higher variability in VADER is not associated to white mental health

Any Bad Days

Regression results examining rolling average standard deviation and any days of bad mental health - White
term estimate std.error statistic p.value
(Intercept) 0.662 0.150 4.406 0.0
X_age_g -0.050 0.000 -160.694 0.0
X_educag -0.015 0.000 -31.304 0.0
female 0.130 0.001 142.474 0.0
INCOME2 -0.001 0.000 -46.803 0.0
MARITAL 0.023 0.000 80.995 0.0
rolling_sd_VADER90 -0.504 0.393 -1.283 0.2

Higher variability in VADER is not associated to white mental health

Frequent Bad Days

Regression results examining rolling average standard deviation and frequent days of bad mental health - White
term estimate std.error statistic p.value
(Intercept) 0.206 0.102 2.026 0.043
X_age_g -0.011 0.000 -51.941 0.000
X_educag -0.035 0.000 -106.624 0.000
female 0.051 0.001 82.768 0.000
INCOME2 0.000 0.000 -24.582 0.000
MARITAL 0.012 0.000 63.192 0.000
rolling_sd_VADER90 0.022 0.265 0.083 0.933

Higher variability in VADER is not associated to white mental health

Latinx Model

Number of Days

Regression results examining rolling average standard deviation and days of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) -4.457 9.317 -0.478 0.632
X_age_g -0.080 0.021 -3.742 0.000
X_educag -0.060 0.024 -2.464 0.014
female 1.032 0.052 19.774 0.000
INCOME2 -0.008 0.001 -7.894 0.000
MARITAL 0.289 0.014 20.110 0.000
rolling_sd_VADER90 20.231 23.867 0.848 0.397

Higher variability in VADER is not associated to Latinx mental health

Any Bad Days

Regression results examining rolling average standard deviation and any days of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 0.660 0.585 1.129 0.259
X_age_g -0.028 0.001 -21.038 0.000
X_educag 0.013 0.002 8.424 0.000
female 0.075 0.003 22.859 0.000
INCOME2 -0.001 0.000 -12.690 0.000
MARITAL 0.021 0.001 23.483 0.000
rolling_sd_VADER90 -0.792 1.498 -0.529 0.597

Higher variability in VADER is not associated to Latinx mental health

Frequent Bad Days

Regression results examining rolling average standard deviation and frequent of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) -0.687 0.385 -1.784 0.074
X_age_g -0.001 0.001 -1.433 0.152
X_educag -0.004 0.001 -4.051 0.000
female 0.035 0.002 16.114 0.000
INCOME2 0.000 0.000 -6.013 0.000
MARITAL 0.009 0.001 14.993 0.000
rolling_sd_VADER90 2.002 0.987 2.029 0.042

*Higher variability in VADER is associated with a significant increase in the people who report frequent bad mental health.

GLM Models

Latino

Number of Days

Regression results examining rolling average standard deviation and days bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) -2.507 0.012 -213.641 0
X_age_g -0.024 0.000 -560.103 0
X_educag -0.017 0.000 -336.855 0
female 0.293 0.000 2780.998 0
INCOME2 -0.002 0.000 -1084.817 0
MARITAL 0.081 0.000 2817.332 0
rolling_sd_HS90 2.893 0.009 328.017 0

In a Poisson regression, the coefficients estimate the log change in the expected count of the dependent variable for a one-unit increase in the predictor. The exponential of the coefficient (exp(β)) gives the multiplicative effect on the count of the dependent variable per unit change in the predictor. The value 18.04 represents the factor by which the expected count of poor mental health days increases for a one-unit increase in the rolling standard deviation of the health score over 90 days. In simpler terms, as the variability of the health score over 90 days increases by one standard deviation unit, the number of poor mental health days is expected to multiply by approximately 18.04 times, holding all other variables constant.

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!
## Warning: glm.fit: algorithm did not converge
## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
Regression results examining rolling average standard deviation and any bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) 3.312569e+15 1502850.599 2204190609 0
X_age_g 5.667256e+13 5401.856 10491312563 0
X_educag -8.395426e+12 6203.680 -1353297723 0
female 1.434399e+12 13221.323 108491305 0
INCOME2 5.104520e+11 243.048 2100209146 0
MARITAL -5.816593e+12 3634.940 -1600189426 0
rolling_sd_HS90 -3.668017e+15 1118600.891 -3279111181 0

Significant decrease in people reporting any bad mental health days when there is more variability, which is counterintuitive. However, this model was not significant in original model.

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!
## Warning: glm.fit: algorithm did not converge
## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
Regression results examining rolling average standard deviation and frequent of bad mental health - Latino
term estimate std.error statistic p.value
(Intercept) -5.619084e+15 1502850.599 -3738950693 0
X_age_g -1.022039e+13 5401.856 -1892014959 0
X_educag -3.239831e+13 6203.680 -5222435062 0
female 2.207102e+14 13221.323 16693501612 0
INCOME2 -2.365667e+12 243.048 -9733326658 0
MARITAL 5.587961e+13 3634.940 15372912133 0
rolling_sd_HS90 3.910447e+15 1118600.891 3495837986 0

Significant association suggested that higher variability is associated with more people reporting frequent bad mental health days.