1

I'm using firearm trace data (available at the state-level) and attempting to compare it to socially disadvantaged areas (census tracts) by the way of regression analysis. Katz and Schnebly (2011) used population characteristics of tracts to measure the relationship between socially disadvantaged areas and gang members and/or gang crime.

I want to be able to measure whether the number of firearms being recovered in a state that are sourced to a different state are dependent on socially disadvantaged areas (certain neighborhoods in urban areas). Assuming that most firearms are being recovered in large urban areas due to higher levels of firearm violence in cities. Since my DV is only at the state-level, is there a meaningful way to generalize tract-level population characteristics to state-level?

I realize that one method could be summarizing these characteristics, but I wasn't sure if there would be any variation lost between multiple urban areas in states.

There is no way to disaggregate the state trace data to counties or tracts.

Firearm Trace Data
Recovered: (Factor) State where firearm was recovered and traced by law enforcement
Year: (Integer) Year when trace request was initiated
Source: (Factor) State where firearm was first purchased
NumWeapons: (Numerical) Number of firearms that were recovered in state (Recovered) and successfully traced to a state (Source) during Year

The socially disadvantaged data consist of index scores (factor analysis using rates of poverty, single-families, 25 and over w/o HS degree, etc.) for each census tract in a state by year.

4
  • Where did you find the data on firearms traced/recovered in a state? Is it open source?
    – csk
    Feb 11, 2019 at 19:25
  • It is open source (ATF, Data and Statistics, Firearms Trace Data 2013-2017: atf.gov/resource-center/data-statistics)
    – James R.
    Feb 12, 2019 at 13:06
  • It seems your question now has two parts: 1. Is there a statistically meaningful way to generalize the tract-level data; and 2. How to do part 1. You need an answer to part 1 before we can help with part 2. This is not the best site for questions about statistics. I recommend stats.SE for part 1.
    – csk
    Feb 12, 2019 at 18:21
  • Thanks for the suggestion, and I posted the same question to them earlier. I'll be sure to update this with an answer as soon as I can.
    – James R.
    Feb 12, 2019 at 18:28

1 Answer 1

2

When you have data sets of different levels of detail, you can only make comparisons at the lowest level of detail. There's no meaningful way to split up the state-level data; you certainly can't assume that the firearms traced/recovered in a given state were evenly distributed around the state. The only thing you can do is generalize the census tract-level data. So the next question is this:

Is it possible to generalize the census tract data to the state level in a meaningful way? This is a better question for the statistics SE site. My gut tells me the answer is no, but I'm not a statistician.

Assuming you do find a meaningful way to generalize the census tract data to the state level, then you can make comparisons between states.

With these data sets, the conclusion would look like this:

  • "States with higher levels of social disadvantage have fewer firearms traced/recovered than states with lower levels of social disadvantage."

-or-

  • "States with higher levels of social disadvantage have more firearms traced/recovered than states with lower levels of social disadvantage."

-or-

  • "States with higher levels of social disadvantage have similar numbers of firearms traced/recovered than states with lower levels of social disadvantage."

In any case, remember that correlation does not imply causation.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.