Tag: Rice et al. 2021 JOP

See here for a discussion of the Rice et al. 2021 mock juror experiment.

My reading of the codebook for the Rice et al. 2021 experiment is that, among other items, the pre-election survey included at least one experiment (UMA303_rand), then a battery of items measuring racism and sexism, and then at least another experiment. Then, among other items, the post-election survey included the CCES Common Content racial resentment and FIRE items, and then the mock juror experiment.

The pre-election battery of items measuring racism and sexism included three racial resentment items, a sexism battery, three stereotypes about Blacks and Whites (laziness, intelligence, and violent), and 0-to-100 feeling thermometers about Whites and about Blacks. In this post, I'll report some analyses of how well these pre-election measures predicted discrimination in the Rice et al. 2021 mock juror experiment.

---

The first plot reports results among White participants who might be expected to have a pro-Black bias. For example, the first estimate is for White participants who had the lowest level of racial resentment. The dark error bars indicate 83.4% confidence intervals, to help compare estimates to each other. The lighter, longer error bars are 95% confidence intervals, which are more appropriate for comparing as estimate to a given number such as zero.

The plotted outcome is whether the participant indicated that the defendant was guilty or not guilty. The -29% for the top estimate indicates that, among White participants who had the lowest level of racial resentment on this index, the percentage that rated the Black defendant guilty was 29 percentage points lower than the percentage that rated the White defendant guilty.

The plot below reports results among White participants who might be expected to have a pro-White bias. The 26% for the top estimate indicates that, among White participants who had the highest level of racial resentment on this index, the percentage that rated the Black defendant guilty was 26 percentage points higher than the percentage that rated the White defendant guilty.

---

The Stata output reports additional results, for the sentence length outcome, and for other predictors: a four-item racial resentment index from the post-election survey, plus individual stereotype items (such as for White participants who rated Blacks higher than Whites on an intelligence scale). Results for the sentence length outcome are reported for all White respondents and, in later analyses, for only those White respondents who indicated that the defendant was guilty.

---

NOTE

1. Data for Rice et al. 2021 from the JOP Dataverse. Original 2018 CCES data for the UMass-A module, which I used in the aforementioned analyses. Stata code. Stata output. Pro-Black plot: dataset and code. Pro-White plot: dataset and code.

Forthcoming at the Journal of Politics is Rice et al. 2021 "Same As It Ever Was? The Impact of Racial Resentment on White Juror Decision-Making".

---

See the prior post describing the mock juror experiment in Rice et al. 2021.

The Rice et al. 2021 team kindly cited my article questioning racial resentment as a valid measure of racial animus. But Rice et al. 2021 interpreted their results as evidence for the validity of racial resentment:

Our results also suggest that racial resentment is a valid measure of racial animus (Jardina and Piston 2019) as it performs exactly as expected in an experimental setting manipulating the race of the defendant.

However, my analyses of the Rice et al. 2021 data indicated that a measure of sexism sorted White participants by their propensity to discriminate for Bradley Schwartz or Jamal Gaines:

I don't think that the evidence in the above plot indicates that sexism is a valid measure of racial animus, so I'm not sure that racial resentment sorting White participants by their propensity to discriminate for Bradley or Jamal means that racial resentment is a valid measure of racial animus, either.

---

I think that the best two arguments against racial resentment as a measure of anti-Black animus are:

[1] Racial resentment on its face plausibly captures non-racial attitudes, and it is not clear that statistical control permits any post-statistical control residual association of racial resentment with an outcome to be interpreted as anti-Black animus, given that racial resentment net of statistical control often predicts outcomes that are not theoretically linked to racial attitudes.

[2] Persons at low levels of racial resentment often disfavor Whites relative to Blacks (as reported in this post and in the Rice et al. 2021 mock juror experiment), so the estimated effect for racial resentment cannot be interpreted as only the effect of anti-Black animus. Racial resentment in these cases appears to sort to low levels of racial resentment a sufficient percentage of respondents who dislike Whites in absolute or at least relative terms, so that indifference to Whites might plausibly be better represented at some location between the ends of the racial resentment measure. But the racial resentment measure does not have a clear indifference point such as 50 on a 0-to-100 feeling thermometer rating, so -- even if argument [1] is addressed so that statistical control isolates the effect of racial attitudes -- it's not clear how racial resentment could be used to accurately estimate the effect of only anti-Black animus.

---

NOTES

1. The sexism measure used responses to the items below, which loaded onto one factor among White participants in the data:

[UMA306bSB] We should do all we can to make sure that women have the same opportunities in society as men.

[UMA306c] We would have fewer problems if we treated men and women more equally.

[UMA306f] Many women are actually  seeking special favors, such as hiring policies that favor them over men, under the guise of asking for "equality."

[UMA306g] Women are too easily offended.

[UMA306h] Men are better suited for politics than are women.

[CC18_422c] When women lose to men in a fair competition, they typically complain about being discriminated against.

[CC18_422d] Feminists are making entirely reasonable demands of men.

Responses to these items loaded onto a different factor:

[UMA306d] Women should be cherished and protected by men.

[UMA306e] Many women have a quality of purity that few men possess.

2. Data for Rice et al. 2021 from the JOP Dataverse. Original 2018 CCES data for the UMass-A module, which I used in the aforementioned analyses. Stata code. Stata output. Data and code for the sexism plot.

3. I plan a follow-up post about how well different measures predicted racial bias in the experiment.

Forthcoming at the Journal of Politics is Rice et al. 2021 "Same As It Ever Was? The Impact of Racial Resentment on White Juror Decision-Making". In contrast to the forthcoming Peyton and Huber 2021 article at the Journal of Politics that I recently blogged about, Rice et al. 2021 reported evidence that racial resentment predicted discrimination among Whites.

---

Rice et al. 2021 concerned a mock juror experiment regarding an 18-year-old starting point guard on his high school basketball team who was accused of criminal battery. Participants indicated whether the defendant was guilty or not guilty and suggested a prison sentence length from 0 to 60 months for the defendant. The experimental manipulation was that the target was randomly assigned to be named Bradley Schwartz or Jamal Gaines.

Section 10 of the Rice et al. 2021 supplementary material has nice plots of the estimated discrimination at given levels of racial resentment, indicating, for the guilty outcome, that White participants at low racial resentment were less likely to indicate that Jamal was guilty compared to Bradley, but that White participants at high racial resentment were more likely to indicate that Jamal was guilty compared to Bradley. Results were similar for the sentence length outcome, but the 95% confidence interval at high racial resentment overlaps zero a bit.

---

The experiment did not detect sufficient evidence of racial bias among White participants as a whole. But what about Black participants? Results indicated a relatively large favoring of Jamal over Bradley among Black participants, in unweighted data (N=41 per condition). For guilt, the bias was 29 percentage points in unweighted analyses, and 33 percentage points in weighted analyses. For sentence length, the bias was 8.7 months in unweighted analyses, and 9.4 months in weighted analyses, relative to a unweighted standard deviation of 16.1 months in sentence length among Black respondents.

Results for the guilty/not guilty outcome:

Results for the mean sentence length outcome:

The p-value was under p=0.05 for my unweighted tests of whether the size of the discrimination among Whites (about 7 percentage points for guilty, about 1.3 months for sentence length) differed from the size of the discrimination among Blacks (about 29 percentage points for guilty, about 8.7 months for sentence length); the inference is the same for weighted analyses. The evidence is even stronger considering that the point estimate of discrimination among Whites was in the pro-Jamal direction and not in the pro-ingroup direction.

---

NOTES

1. Data for Rice et al. 2021 from the JOP Dataverse. Original 2018 CCES data for the UMass-A module, which I used in the aforementioned analyses. Stata code. Stata output. "Guilty" plot: data and R code. "Sentence length" plot: data and R code.

2. I plan to publish a follow-up post about evidence for validity of racial resentment from the Rice et al. 2021 results, plus a follow-up post about how well different measures predicted racial bias in the experiment.