Background Checks for Firearms Purchases

 

Source: National Instant Criminal Background Check Database. (as of 8/2024).

I've been noodling around, looking for state and local measures of firearms ownership. There are polling results from BRFSS, Gallup, etc. There are hunting license data. And here, are instnat background checks for firearms transfers. A couple of caveats right off the bat: Not every background check results in the purchase of a firearm. And multiple firearms can be transferred on a single background check. And, the scope of situations where a background check is required has expanded over the years.

Even given those caveats, it is clear that there are annual patterns, with a small spike in the Spring, and  a larger spike in the Fall, in most years. There is also clearly a huge increase over time, which would be hard to explain by any expansion of circumstances where a background check is required (although that's certainly part of the picture). And, there are clearly spikes well above the annual cycle and the overall trend, including the Fall of 2008 to Spring of 2008, Fall of 2012 to early 2013, the Fall of 2016 to Spring of 2017, and from Spring 2020 to the end of 2021.

Reflecting on a recent analysis

I've been thinking a lot lately. I guess I always think a lot. The beginning of the semester is always a good time to take a step back to look at the bigger picture.

This post will be a <reflection on / reflection of> an analysis that still feels fresh in my head, which I worked on with a few passionate students, and they got published in 2022 in LGBT Health: Walubita T, Beccia AL, Boama-Nyarko E, Ding EY, Ferrucci KA, Jesdale BM. Complicating Narratives of Sexual Minority Mental Health: An Intersectional Analysis of Frequent Mental Distress at the Intersection of Sexual Orientation, Gender Identity, and Race/Ethnicity. LGBT Health. 2022 Apr;9(3):161-168. doi: 10.1089/lgbt.2021.0099. Epub 2022 Feb 17. PMID: 35180360.

I'm thinking of the paper now because it's assigned reading for my social epidemiology class - I always like to include the work of recent students in the syllabus, and I super proud of these particular students. The real focus is about developing an intersectional approach to considering gender identity, sexual orientation, and race/ethnicity simultaneously with a mental health outcome. "Frequent mental distress" is really just a convenient health outcome.

In this blog, I'm planning to run similar analyses to this paper, using about twice as much data from BRFSS.

 

I've also been thinking about "reproducibility" - apparently there's a "reproducibility crisis" in biomedical sciences, including epidemiology, which we have imported from psychology, who have really given the phenomenon a lot more thought than we have. I've been somewhat skeptical that "reproducibility" is really what the "crisis" is about. Often when I hear it come up, it's typically when one study is statistically significant, and another on essentially the same topic (and often in the same dataset), is not. To me, this sounds like vesting too much faith in statistical significance, and too little creativity about imagining a multiplicity of ways to answer a given question.

I'm used to seeing results that aren't identical, and long ago I let go of the idea that there is one canonical answer to any research question. We're used to teaching about (and thinking about) a "target" with "truth" at the center. Which is an extremely valuable metaphor for distinguishing systematic from random error. But I wonder if a better metaphor for thinking about answers to research questions is thinking of a flock of birds or a school of fish. We should be looking at the range of answers, not just assessing the exact "center", and recognize that the answers to any research question are historically contingent, constantly under "flight". I'm not arguing that there is no such thing as "truth", but that truth depends to some degree, lesser or greater, on context - including the context of who is looking for it and how.

So, that's why I've decided to call the process I'm undertaking here a "reflection" rather than an attempt at "replication" or "reproduction". But enough of the philosophical mumbo-jumbo. What did the earlier analysis show, and what does this "update" show?

From the original paper, this chart shows the predicted prevalence of frequent mental distress (14+ days in the last month with poor mental health), by sexual orientation (heterosexual, bisexual, gay/lesbian), race/ethnicity (White, Black, Hispanic/Latine), and gender identity (cismale, cisfemale, transmale, transfemale, gender non-conforming), resulting in 45 groups. I'm not going to get into the modeling approach in detail, but it does adjust for age, and survey year. It was not able to account for weighting to reflect differential selection probabilities. It uses a Bayesian approach to "borrow" information from similar groups to produce estimates that are less influenced by random variation.

The group with the lowest proportion of frequent mental distress was White cisgender heterosexual men, at 8.2%, with the highest being among Hispanic gay/lesbian gender non-conforming people (47.6%), White bisexual transmen (44.3%), and White bisexual gender non-conforming people (44.1%).

The major takeaway I take from this graph is that gender minority and sexual minority respondents were much more likely to have frequent mental distress. Nearly all groups of bisexual people had higher levels of frequent mental distress than any group of heterosexuals. And, there's a lot of heterogeneity that would be poorly explained by thinking about intersectionality as a simple game of summing up departures from the straight White cismale "norm".

The Reflection Analysis

So, this initial analysis was performed in 1,024,261 BRFSS respondents from 2014-2019 (about 35-40 states), in 3 racial/ethnic groups, 3 sexual orientation groups, and 5 gender identity groups. Since the 2023 BRFSS data came out a week ago, there's a lot more data: 2,180,997 respondents from 2014-2023 (all but 3 states: NE, NH, SD), so I've expanded the new analysis to include more racial/ethnic groups (Hispanics of any race(s), non-Hispanic and White only, Black only, American Indian/Alaska Native only, Asian only, Pacific Islander only, another race only, and multiracial), and one more sexual orientation group "something else". I've looked closely into the "something else" group, and it is challenging to interpret, because it is a very heterogeneous mixture of people who prefer another sexual orientation identity (e.g. pansexual, asexual, gray), people who are straight but don't realize they have a sexual orientation, and people who react against the premise of identifying themselves. So, it is a group that has more young people, and more older people in it than other sexual minority groups, more people with limited education, and those who speak languages other than English. The gender identity groups remain the same. Instead of 3x3x5 = 45 groups in the paper, this results in 8x4x5=160 groups in this reflection analysis. The paper took a simpler approach to pruning these groups; in this reflection analysis, I'm excluding groups with fewer than 25 respondents, and also those with a standard error estimate around the crude prevalence of higher than 10%, resulting in 95 interpretable groups.

Also, I don't have the proper training to use a MAIHDA model, so I've done the best I can with looking at weighted, but unadjusted, prevalences, and then model results from a simple linear model that sticklers will certainly take issue with (it is modeling a continuous normal distribution for a "0"/"1" outcome, for instance), but at least it can account for weighting, adjustment for age group and survey year, and I also put in state of residence. 

But let's start with unadjusted results first. These account for weighting, but no covariates.

 

 Overall a similar picture to before, with gender minority and sexual minority respondents more likely to report frequent mental distress, and again, the bisexual group are routinely higher than heterosexuals of any gender. But these results are only for non-Hispanic Whites, and are not adjusted for age - likely to be an important confounder since frequent mental distress tends to be higher in younger age groups, and so does identifying as a gender minority, and to a lesser extent, as a sexual minority. I debated whether to go through the unadjusted analyses for all racial/ethnic groups first, or do some adjustments first. It's a tough call, but I think the adjustment for age is crucial, so I'm going to start there.

These next charts are a bit different - these are showing the difference in prevalence in frequent mental distress from White straight cis men, after adjustment for age group, survey year, and state of residence.