The study that started the modern academic discussion about gun control is the famous 1995 Garry Kleck study. It was a telephone survey that interviewed 5'000 arbitrarily chosen people and estimated that there are around 2.5 million defensive gun uses (DGUs) in the USA each year. This study has been replicated dozens of times, each time with similar results: there are on the order of magnitude of a million defensive gun uses per year in the USA.

However, there is another methodology for determining how many DGUs there are per year: ask non-anonymously people who reported being victimized what they did in order to protect themselves. That's what the NCVS (National Crime Victimization Survey) study is doing, and it estimated that there are around 100'000 defensive gun uses per year.

Today, most social scientists believe the NCVS study is telling the truth. I am having trouble understanding why. It's obvious how the NCVS study could be a giant underestimate: perhaps most defensive gun uses don't get reported to the police, perhaps most people who used a gun to protect themselves don't want to admit that in a non-anonymous survey (out of fear they've done something illegal)... But it's not at all obvious how could the Gary-Kleck-like studies be a giant overestimate. How could they be?

Do most social scientists today believe that Gary-Kleck-like studies suffer from massive telescoping? That people are misremembering events that occurred 8 years ago as if they occurred less than a year ago? Or do most social scientists today perhaps believe that many people are dreaming that they used a gun defensively and are mistaking those dreams for reality? Or something else?

  • 2
    I find it hard to take this as a good-faith question when it says: "Or do most social scientists today perhaps believe that many people are dreaming that they used a gun defensively and are mistaking those dreams for reality?"
    – Obie 2.0
    Commented Jun 24 at 18:48
  • 2
    The two biggest issues that I am aware of are ambiguous terminology around defensive gun use, wherein many people seem to include showing a gun or mentioning that they have a gun—even, possibly, if there was no actual confirmed threat to them and sample size issues—only 56 people reported defensive gun use, so the upper bound of the 95% confidence interval on the population proportion is a full 70% higher than the lower bound (or over 100% higher for the 99% confidence interval).
    – Obie 2.0
    Commented Jun 24 at 18:59
  • 2
    That's from purely random variation, but the small percentage leads to even worse possible issues with non-random variation (bias): namely, dishonesty on surveys. It may well be that very few people would lie on that survey. Maybe only, say, 1% of the population would claim to have used a gun defensively just as a joke. But wait, that's on the same order of magnitude as the reported effect size!
    – Obie 2.0
    Commented Jun 24 at 19:04
  • 5
    If you ask people in an anonymous telephone survey who big their dick is, do you believe you will get an accurate estimate?
    – quarague
    Commented Jun 24 at 19:28
  • 3
    @Obie2.0 I think they meant more like people are making assumptions of others' motives/intents. Imagine Alice walks through a bad neighborhood. She sees Bob loitering on her intended path. When Bob looks her way, she thinks she's being sized-up in preparation for an attack. Alice puts her hand in her pocket on her gun, and believes Bob knows this. Alice walks by, and Bob does nothing. Alice may count this as a DGU, believing Bob was dissuaded by her being obviously (to her) armed. But if you asked Bob, he never planned on attacking her and didn't know she was armed.
    – Ryan_L
    Commented Jun 24 at 21:07

1 Answer 1


Not necessarily in order of importance

Polling for low probability events is challenging. If 2.5 million events per year is the right number, that would mean that something like 1% of adult-ish people use a gun defensively every year (it is implausible that 5 year olds are using guns defensively but plausible that 15 year olds are hence "adult-ish"). Even if you poll thousands of people, you're only going to get tens of positive responses. That means that the 95% confidence interval is very large (the uncertainty in the results of the poll is great) and any biases are amplified. Respondents that have used a gun defensively may be particularly likely to complete the survey. Respondents may report older events or events experienced by family and friends because they want those events counted. And something north of 1% of survey respondents are likely to not take the survey seriously and make up answers because they enjoy causing chaos.

Defensive gun use may be subjective. If a someone walks up to a house, the homeowner calls out that they are armed, the person runs off, and no police report is ever made, the situation is pretty ambiguous. The homeowner might believe they have used their gun defensively to scare off a would-be burglar. If the person walking up to the house was a door-to-door salesman rather than a would-be burglar, however, he'll have a rather different story about a menacing gunowner. If a small number of people are watching a lot of true crime TV and are convinced that kidnappers are lurking at their local Target, they could account for a large fraction of reported defensive gun uses by scaring away perfectly innocent shoppers.

And, of course, there are arguments about different sorts of methodology errors in any given paper. There are arguments that individuals in particular regions were oversampled which could skew the results when extrapolated nationally. Some of the earliest surveys could have picked up individuals who used guns defensively in their work (i.e. police, security guards) or against animals. Subsequent surveys done by the same author taking in to account many of these criticisms have produced lower estimates (though still more than 1 million annual DGUs).

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .