Deltoid

Dr. Paul H. Blackman writes:

I was curious about the suggestion that hardly anyone could possibly still believe the Kleck data now that NSPOF had become the 15th or 16th such survey in the same general category.

Then you seem to have misunderstood. Kleck’s estimate (not his data - I have no problem with his data, just his interpretation of it) is not credible because it fails every single cross check of its validity.

It is inconsistent with:

CDC counts of homicides
UCR counts of homicides
Kleck’s own, earlier, estimate of defensive woundings
Kleck’s own, earlier, estimates of defensive killings
Wright and Rossi’s survey of criminals
NEISS counts of gunshot wounds
NCVS counts of burglaries and violent crimes
UCR counts of burglaries and violent crimes
NCVS counts of DGUs
Dade county study of gun use by CCW holders.
Kellermann’s Atlanta study of gun use against intruders.
NCVS and UCR counts of gun crimes

His data also contains anomalously high percentages of defences by the respondent and by women.

The only thing that it agrees with are other surveys which are just as vulnerable to fabrications by a small percentage of respondents.

As for the NSPOF survey, it has similar problems. Cook notes that “almost half the incidents appear to contain some internal inconsistency, or otherwise do not make sense.”

Guns can be used against criminals perceived to be using guns (roughly one million such crimes would be projected lately from NCVS)

One million? But in his paper Kleck gives 550,000 as a “generous” estimate. Is the NCVS right and Kleck wrong?

and against criminals not perceived to be using guns, and thus can easily outnumber gun-related crimes.

Have you considered the possibility that an armed criminal might encounter an unarmed victim? Some folks have even suggested that they might seek such victims out.

And Kleck would be the first to note that the various breakdowns — say 18% of protective gun use — are less reliable than the overall figure,

A 95% confidence interval is 300,000 to 600,000. The statistical reliability of this figure is similar to that of the Hart poll and much more than that of the Mauser poll. It is odd that Kleck has not discounted these estimates too.

In any case, even if we use the low end of the interval and the NCVS estimate of criminal gun use instead of the Kleck one, we still find the unlikely estimate that 30% of gun crimes involve the victim pulling a gun on the perp.

NCVS remains the sport, and understandably since it’s the survey not aimed at measuring protective use of anything.

This is untrue. The questions have been posted here and it is quite clear that are interested in finding out what crime victims do for protection. Kleck even uses it for that purpose.

What is true is that most of the surveys that agree with Kleck were not designed for measuring protective uses. It is odd that you do not discount these surveys because of this.

I don’t think Killias and others were afraid of one particular “killer question” so much as a series of questions regarding the limitations of the studies on which they would be relying.

OK, then share this series of questions with us. If such a series of questions exists then you and Kleck would surely want to give it wide publicity.

I don’t know why Killias and some other anti-gun researcher bowed out, but they should have known in advance that testimony would involve a trip to Canada, but may not have known in advance that it would involve Kleck’s aid in the crossexamination to follow any testimony. Why, after all, provide an affidavit for submission if that affidavit cannot be submitted unless in-person testimony also occurs?

That isn’t the way things worked when I submitted an affidavit as an expert witness to a court. I was told that I only needed to appear if the other side wanted to cross examine me. Nor was the court date scheduled for my convenience.

Incidentally, using an unpublished CDC study involving 36 generally affluent nations, with gun use in suicide as the surrogate measure for gun availability, Kleck’s reanalysis indicates that with or without the U.S. excluded, “there is no significant (at the 5% level) association between gun ownership levels and the total homicide rate in the largest sample of nations available to study this topic.”

Ahem. What he probably found was that there was no significant correlation between the gun suicide rate and the the total homicide rate. That’s also true for Killias’ data set. But if you use a survey based measure of gun ownership as Killias did, you do find some significant correlations.

That is, Kleck got a different result because he used a surrogate measure rather than a direct one.

“Eugene Volokh” writes:

but I was wondering what you thought about the NCVS point I raised again a few days ago. To my knowledge, waiting for respondents to volunteer information is generally considered rather bad survey practice; and we saw that with the rape statistics shifting to a direct question changed the total by about a factor of 2.5 or 3, if I recall correctly.

I have even been told — entirely outside the defensive gun use context — that the trick is cuing as often as possible: Asking the

question directly, several times, in subtly different ways, to trigger people’s memories (and perhaps willingness to respond).

This was considered in the NCVS redesign — it asks the screening questions in several different ways. This would seem to be better than just having one screening question, as Kleck’s survey does.

How exactly is this done in the NCVS redesign? I see a good deal of extra cuing as to rape, but little as to defensive use.
There are indeed two questions introducing the main issue: “Did you do anything with the idea of protecting YOURSELF or your PROPERTY while the incident was going on?” and “Was there antything you did or tried to do about the incident while it was going on?”

I was refering to cueing to get the person to recall the crime incident, not the detail of using a gun for defence. Because it uses more cueing questions, you would expect the NCVS to be better at getting the person to recall the incident in which the DGU occured than Kleck’s survey.

But if someone says “Oh, I shouted at the guy and he ran away,” there’s to my knowledge NO follow-up question “What did you shout?” or “Did you shout anything about a weapon?”

I think they just ask something like “Anything else?”

Now of course some people will say “I shouted `I’ve got the gun’ and he ran away.” But others won’t be that specific, and not just because they forgot about the gun or are reluctant to talk about it.
Some people, in an interview like this, will go into gory detail with the mildest of prompting. Others, and this is often just a matter of temperament, will give a relatively short answer, especially when it’s an answer to question #42 (I realize it might not actually be the 42nd question — some might have been skipped — but it’s not the second, either) and no end obviously in sight. If you don’t ask for more detail, you won’t get it. This is the lack of cuing that strikes me as particularly problematic, though I agree that both forgetfulness and reluctance are troublesome, too.

I dunno. It seems to me that stating “I shouted at the guy and he ran away,” when you in fact used a gun for defence is highly misleading and that most people are not so poor communicators that they would do this accidently. The only way to find out for sure is to experiment with different questions. I don’t know if this is one of the things they tried out in the NCVS redesign.

Finally, though, I’m happy to hear that there seems to be some agreement that the NCVS probably undercounts, at least by a factor of two (though I wonder why it would be just a factor of two).

I didn’t say at least, I said at most. Since Kleck turned up so many fabricated DGUs, it seems probable that some of the NCVS DGUs were also fabrications. If this overcounting exceeds the undercounting we have already discussed then the NCVS will overestimate the number of DGUs. Even if the two factors cancel out and the NCVS estimate is correct it means that you cannot use the NCVS to argue that guns are the most effective means of self-defence.

Why a factor of two? It seems to me that a majority of gun owners will not answer straightforward questions in a misleading way. (If exactly half did, then the NCVS would bu out by a factor of two.) Also, the indirect estimates I have been to make all seem to end up as the same order of magnitude as 80,000.

Do we agree, then, that saying “there are 80,000 defensive gun uses a year” is inaccurate (just as I personally agree that saying “there are 2.5 million defensive gun uses a year” is inaccurate — I always try to qualify any such statements about total uses)?

I’ve certainly not made such a statement. I would say “The most reliable estimate of DGUs is the NCVS one of 80,000, but even this might be out by a factor of two.”

Peter Boucher writes:

Just in case anyone’s interested.

Copied from Kleck/Gertz, here are the polls from table 1 (minus those with no estimate of annual DGUs):

Survey, Where, What year, What kinds of guns, # DGUs

Field, California, 1976, just handguns, 3.1M
Bordua, Illinois, 1977, all guns, 1.4M
DMIa, U.S., 1978, all guns, 2.1M
DMIb, U.S., 1978, all guns, 1.1M
Hart, U.S., 1981, just handguns, 1.8M
Ohio, Ohio, 1982, just handguns, 0.8M
Mauser, U.S., 1990, all guns, 1.5M
Gallup, U.S., 1991, all guns, 0.8M
Gallup, U.S., 1993, all guns, 1.6M
L.A.Times, U.S., 1994, all guns, 3.6M
Tarrance, U.S., 1994, all guns, 0.8M

I’ve just spent a couple of hours figuring out how Kleck computed these estimates. Kleck acknowledges that all these polls have deficiencies when used to attempt to estimate DGUs, so he applies various correction factors for these deficiencies to construct these estimates. For example, the Ohio poll used a recall period of “ever”, only asked about handguns, and only asked respondents in handgun households. So Kleck took the 6.5% who said they had used a gun, multiplied it by the adult population of the US (190M), then multiplied by 0.215 to correct for the question only being asked of handgun households (I guess that he is estimationg that 21.5% of US households have handguns), then multipled by 1.21 to correct for only handgun uses being counted (1.21 is the ratio between all gun crimes and gun crimes - Kleck seems to be assuming that the ratio will be similar for DGUs) and finally multiplied by 0.237 to adjust the number from “ever used” to “used in past year” (I don’t know where this number comes from - it seems way too high) to get an estimate of 0.8M.

These correction factors are rather arbitrary and different choices can give wildly different results. For example, in “Point Blank”, because excluding defences against animals caused the percentage who used a gun to drop from 12% to 7% in the DMIb poll, Kleck used a correction factor of 7/12 to correct polls that did not exclude uses against animals. In the Kleck/Gertz paper, this factor is not corrected for. Another example: Since the average gun owner has had their guns for an average of 20 years, a reasonable way to convert from a poll that asked if the respondent had “ever used” to get uses per year, would be to apply a correction factor of 1/20, rather than Kleck’s 0.237. Just make a different choice on these two factors causes the estimates derived from some polls to come out in agreement with the NCVS, rather than Kleck. For example, the estimate you get from the Bordua poll is about 150k — much closer to the NCVS than Kleck.

So, it cannot be said that all these polls support Kleck rather tan the NCVS.

There is another very interesting thing about the numbers in Table 1 — we can use them to test my hypothesis that a large number of the gun uses are fabrications.

Some polls asked about handgun uses while others asked about all gun uses. Now, since it is just as easy to make up a handgun use as any sort of gun use, I would expect the percentage who used to be about the same, no matter whether the question was asked about handguns or any sort of gun. On the other hand, Kleck would expect the all guns polls to give higher usage percentages by a factor of roughly 1.21. (Recall that’s the correction factor he uses for handgun-only polls).

So, I looked through table 1 and compared all the pairs of polls that differed only in the handgun/any gun-use question. (That is, polls that used the same recall period and so on.) For each pair I calculated the ratio (all-gun poll)/(handgun poll). I expect this ratio to be about 1, Kleck expects it to be 1.21.

The results:

Poll pair       Ratio
Field-Bordua    5/8.6  = 0.6
Hart-Mauser     3.79/4 = 0.9
Hart-Tarrance   2/4    = 0.5
Hart-Kleck      3.898/4= 1.0
Ohio-Gallup91   8/6.5  = 1.2
Field-LA Times  8/8.6  = 0.9
        Mean ratio = 0.9

Of the six pairs, five come out closer to my 1.0 and one closer to Kleck’s 1.21.

Hence we can see that the polls in Kleck’s table 1 show evidence that the respondents have fabricated many of the DGUs reported.

SFBearCop wrote:

I can think of a number of reasons, none of them noble, why someone would fabricate a DGU, starting with giving the pollster what they thought was wanted. People do it all the time, so a friend in the public-opinion-counting game told me thirty or more years ago.

John Briggs writes:

This would account for some false positives in the DGU surveys. It would also be present, presumably, in NCVS responses. The question is why are the response rates so different?

The DGU question appears quite early in Kleck`s survey. It’s not hard for a person to guess that it is the important question and give the interviewer what they think the interviewer wants. With the NCVS it occurs after many other questions and without the clue of mentioning a gun in the question.

Does the relative lack of interest the NCVS shows in defensive behavior in general prompt false negatives and false silences that affect its reported DGU rates?

Quite possibly. But a false positive rate of 2% gives you two million bogus DGUs, while a false negative rate as high as 50% means you only underestimate by a factor of two.

SFBearCop wrote:

Follow that with the need to seem more important than you are. It’s called bragging. Most men, and a few women, don’t like to be thought lacking in the right stuff. Of course they used a gun to deal with some humna varmint. It’s in the traditions of our nation, of western history, of all we hold sacred!

John Briggs writes:

Certainly a factor. But would it not be more manly to use one’s bare hands?

Perhaps. But the structure of Kleck’s survey doesn’t give you that option. If you want to brag to Kleck’s surveyor, you have to make up a DGU.

What we don’t know is the degree of bragging. Why don’t respondents brag to NCVS interviewers to the same degree?

If you want to brag how you thwarted a crime to the NCVS you don’t have to claim you used a gun. You can be more manly and say that you used your bare hands ;-) Also, the NCVS interviews the same household every six months over a three year period. The first interview turns up about 50% more crime than subsequent ones. THe NCVS believes this is caused by “telescoping” and discards the first interview. Some of the difference may be caused by braggers who give up bragging when they find that it doesn’t impress the NCVS interviewer.

Another reason why people might lie:

The respondent could be afraid that the interviewer is really a criminal “casing the joint”, so she makes a DGU to make it clear that she has a gun and is prepared to use it against any criminal that breaks in.

John Briggs writes:

[Calculation of number of justifiable shootings deleted] This would suggest 15,000 to 20,000 civilian justifiable woundings or 17,500 to 22,500 incidents in which a civilian shot and hit an assailant.

Kleck does a similar calculation in “Point Blank” to get an estimate of 10,000 to 20,000.

(This represents an awfully high figure if there are only 80,000 civilian DGUs as the NCVS reports–of course, the NCVS could be low.)

As you have noted, if we know A, the fraction of DGUs where the defender shot at the criminal, and B, the fraction of DGUs where one or more of the shots fired at the criminal actually hit, then we could estimate the number of DGUs. Unfortunately, the only information about what the value of A is comes from the NCVS and Kleck’s survey and if we are going to trust either one, we might as well just use the direct estimate of DGUs that it produces.

Furthermore, we don’t know B that well either. Kleck gives figures of 37% for police and 18% for criminals (p173), but it might be different for civilian DGUs. So, I’ll use both NCVS and the Kleck survey to estimate B and see if the results are reasonable.

NCVS: A=0.4, so B=(10,000 to 20,000)/(0.4*80,000)=30 to 60%
Kleck: A=0.16 to 0.24, so B=(10,000 to 20,000) /((0.16 to 0.24)*2,500,000) = 2 to 5%

The top of the NCVS range for B seems rather high, but 30% seems like reasonable number for B. Kleck’s survey gives numbers for B that are way too low.

Viktor writes:

Of course, we know here in America that the highest crime rates for the past 50 years are in the cities that have the strictest gun control laws (Washington, D.C., New York City, Chicago, Detroit) imposed on innocent people - the distinction being, there can be no gun control that is effective on criminals since they are willing to break the law in the first place.

Too bad for you that it easy to find the actual statistics on the web.

Aggravated Assault

OFFICIAL DEFINITION: An unlawful attack by one person pon another for the purpose of inflicting severe or aggravated bodily injury. This type of assault usually is acccompanied by the use of a weapon or by means likely to produce death or great bodily harm. Simple assaults Are excluded.

RANK CITY
 1 Baton Rouge, LA
 2 Miami, FL
 3 Atlanta, GA
 4 Tampa, FL
 5 Newark, NJ
 6 Peoria, IL
 7 Flint, MI
 8 St. Louis, MO
 9 Orlando, FL
 10 St. Petersburg, FL
 12 Chicago, IL
 15 Detroit, MI
 24 Los Angeles, CA
 40 Dallas, TX
 61 Houston, TX
 72 New York, NY
 77 Phoenix, AZ
 80 San Diego, CA
 130 Philadelphia, PA
 187 San Antonio, TX

None of the four cities you cite are in the top ten for aggravated assault. Four of the top ten are in Florida, where it is legal to carry concealed weapons.

It is certainly no statistical anomaly that the highest crime is in the areas with the most extreme gun control laws restricting the innocent from self-defense.

What? Like Florida?

John Briggs writes:

[Calculation of number of justifiable shootings deleted] This would suggest 15,000 to 20,000 civilian justifiable woundings or 17,500 to 22,500 incidents in which a civilian shot and hit an assailant.

Kleck does a similar calculation in “Point Blank” to get an estimate of 10,000 to 20,000.

For reasons I allude to below I am inclined to believe that civilian DGUs would be likely to result in a significantly lower killed-to-wounded ratio than would criminal gun use. The 10% to 15% lethality ratio of gunshots may lump together much higher kill-ratio criminal shootings intended to kill with a much lower lethality ratio for civilian DGUs that would, I think, more often include semi-warning shots likely to wound rather than kill. Civilian DGUs do not have as their specific intent the killing of their target so I am comfortable with the higher figure for woundings.

I think you’ll find that most criminal shootings are not unambiguous attempts to kill (they generally do not finish the victim off with a head shot at point blank range.) Some criminal shootings are deliberate woundings. A large number of criminal shootings are “drive-bys” — fired from long range and more likely to hit an extremity than a self-defence shooting at close range. These factors suggest that defensive shootings would be more lethal than criminal ones.

On the other hand, criminal shootings include some execution style shootings, and you would expect that a self-dence shooter would be more likely to call an ambulance. These factors suggest that defensive shootings would be less lethal than criminal ones.

Overall, I don’t think you can come to a conclusion either way.

And, now for an inchoate thought: would not the propensity of braggarts to fabricate DGUs be immune to telescoping and forgetfulness, unlike honest respondents who would be depending on memories of actual events? If I am correct in recalling that Kleck’s five-year data suggested only 800,000 DGUs, this substantial fall-off compared to the 2,500,000 DGUs based on one-year positives seems to argue against fabrications as a significant factor since there seems no reason that braggarts would prefer to fabricate DGUs only in the one-year time period. Their influence would seem to be substantially outweighed by honest (if not always accurate) respondents whose recall would be subject to the conflicting factors of forgetfulness in the long run and telescoping in the short.

The enormous difference between 800,000 (5 year recall, houshold DGUs) and 2.5M (1 year recall, personal DGUs) is in fact string evidence that there is some fabricating going on. Suppose you have made up a DGU. Now Kleck asks you for details. “Was it within the past year?” You answer randomly. So, you would expect about 50% of the fabricated DGUs to occur in the past year. On the other hand, a real DGU would be equally likely to fall into any of the five years and you would expect about 20% to occur in the past year. In fact Kleck found 40% said that it had occured in the past year, which suggests that there is more fabricating than truth-telling going on. Because 40% is twice 20%, you get a factor of two difference in the estimate depending on whether you use a 5 year recall or a one year recall. Kleck explanation for this disrepency is recall failure - that is, half of DG users forget the incident after a couple of years.

In trying to resolve the discrepancies between Kleck and the NCVS regarding the percentage of “shots-fired” incidents I wonder if the differences lie, in part, in the the focus of the latter on commission of a crime and some sort of “legal” victimization status for the respondent. Perhaps the survey process suppresses reports of numerous actual but borderline cases. The NCVS respondents were, perhaps, much farther along in the crime victimization process (so they were more obviously in danger and prompted to shoot) whereas Kleck’s respondents include not only NCVS types but, say, the female apartment dweller who refers to her handgun to send a persistent stranger away from her door.

Kleck take pains to only count DGUs against actual crimes. I suppose it is possible for this sort of incident to be included in Kleck’s study if the respondent lied and said (for example) that the the stranger tried to break in. (Presumably to make her actions seem more reasonable. )

Furthermore, we don’t know B that well either. Kleck gives figures of 37% for police and 18% for criminals (p173), but it might be different for civilian DGUs. So, I’ll use both NCVS and the Kleck survey to estimate B and see if the results are reasonable.

NCVS: A=0.4, so B=(10,000 to 20,000)/(0.4×80,000)=30 to 60%
Kleck: A=0.16 to 0.24, so B=(10,000 to 20,000) /((0.16 to 0.24)x2,500,000) = 2 to 5%

The top of the NCVS range for B seems rather high, but 30% seems like reasonable number for B. Kleck’s survey gives numbers for B that are way too low.

But the reasonableness of the 30% figure lies (only?) in the fact that it is between the police 37% and the criminal 18%, yes? And the unreasonableness of Kleck’s 2% to 5% derives (only?) from the fact that it is so much lower than the criminals’ 18%, yes? Actually, it derives partly from use of Kleck’s one-year-data estimate rather than his five-year estimate. Preliminarily he suggested the longer-term data would give an estimate of 800,000 DGUs each year but he preferred the one-year figure because he felt forgetfulness was more of a problem than telescoping. Does the final published version note the lower five-year estimate?

Afraid not. The lowest estimate he has is 1.2M, and it’s clear that he regards the 2.5M one as best.

plug in 800,000 in place of 2,500,000 we get:

Kleck (against his will): B = 6.5% to 15.6%

I realize his 2,500,000 number gets the publicity and, apparently, he considers it the better of the two numbers but is there any data on the relative impacts of forgetfulness vs. telescoping to support his preference for the one-year data?

He cites some data that suggests that telescoping would make no more than a 20% difference. So the furthest you can go here is to get up to almost 6%, whiich still seems unlikely.

In any event, your calculation should prompt us to wonder how civilian DGUs ought to compare with police and criminal “success” rates. If the inclinations of the civilian and the characteristics of his situation are markedly different from those of police and criminals we might well find that the percentage of situations in which a civilian “shots-fired” incident resulted in a “hits-made” incident differs greatly from the performance of police and criminals.

For example, a police officer is under a duty to stay engaged with a criminal to apprehend him even if he flees and even at the risk of prolonging or re-starting an exchange of gunfire. A civilian, in contrast, would count it a success either to prompt his assailant to flee or to disrupt his assailant’s attack sufficiently to allow himself to retreat. Civilian DGUs ought to involve fewer shots, on average.

Not necessarily. The attacker is probably more motivated to attack his victim than to fight third parties (like police).

The general inaccuracy of gunfire under stress suggests to me that there would be many more short exchanges of gunfire, characteristic of a civilian DGU, that resulted in no one getting hurt than there would be with police against criminal gunfire. A criminal’s use of gunfire would also, I think, be significantly more likely to result in his target getting hit because the criminal more often specifically wants his victim dead either because that is the sole purpose of the shooting or because he fears retaliation from his victim or because he wishes to leave no witnesses to his “main” crime. He is less likely to quit shooting until he hits his victim or he runs out of ammunition. These factors would tend to drive the criminal B ratio above that of the civilian.

On the other hand criminal shootings are more likely to be at long range where the chance of shootings are less.

While I’m here, this is my estimate of the number of DGU-related woundings.

The NEISS estimates 60k Assault/legal intervention nonfatal firearm injuries treated in hospitals. According to the NCVS about 90% of those wounded in this way get hospital treatment, so there are roughly 70k such woundings each year. Add in 18k deaths and I estimate about 90k assault/legal intervention shootings altogether. Using the data from Table 4.2 (on Civilian Legal Defensive Homicides) as Kleck does, about 10-13% of the woundings will be self-defence, or 9k-12k, i.e roughly 10,000.

Alan Peyton-Smith writes:

Oh, BTW, I have good reason to believe that this Brian Ross we’re talking about is the same Brian Ross who created a few extra Internet accounts for himself under false names such as David Bowman and Kylie Minogue and Tim Lambert and possibly several others.

Are you really gullible enough to believe everything you read on Usenet? A gentleman who goes by the name of Nosy has been making the absurd allegation that I am really a pseudonym for Brian Ross. He’s doing this in a rather lame attempt to annoy me.

I’ll be happy to prove that I’m not a pseudonym fro Brian Ross right after you do. I mean, isn`t it a bit a suspicious that you ALWAYS get comprehensively trounced whenever you argue with him? Surely a real person could occasionally win one? Could it be that Brian has invented you to make himself look good? Well?

I have on several occasions asked Brian to deny that he has ever engaged in this sort of bizzare behaviour, he has never replied.

Well I deny that I am Brian Ross. If you don’t believe me, you must concede that Brian has denied your bizarre allegation.

A large number of criminal shootings are “drive-bys” — fired from long range and more likely to hit an extremity than a self-defence shooting at close range. These factors suggest that defensive shootings would be more lethal than criminal ones.

John Briggs writes:

Any data on the proportions of such long range shootings? I confess I have not seen a serious treatment of the topic. News acounts leave one with the impression that such shootings involve whole carloads of machinegun equipped gangbangers. Our streets aren’t that wide here in the US. We aren’t talking 100 yard firefights. Whether the extra few yards of range offsets the more serious intent of the shooters I cannot say.

By long range I meant, of course, the width of the street.

There was a study of gunshot wounds treated in an inner-city hospital published in J of Trauma a couple of years ago. They found that on average shotgun wounds (mainly bird shot) were less serious than handgun wounds. This suggests that they are not doing the shooting at really close range.

The enormous difference between 800,000 (5 year recall, houshold DGUs) and 2.5M (1 year recall, personal DGUs) is in fact string evidence that there is some fabricating going on. Suppose you have made up a DGU. Now Kleck asks you for details. “Was it within the past year?” You answer randomly. So, you would expect about 50% of the fabricated DGUs to occur in the past year. On the other hand, a real DGU would

Is that the order of the questioning? “Ever had a DGU? Past year? Past five years?”

No, the first question asks if they had one within the last five years. If they say yes, he asks if it was within the past year.

OK, assume that indeed half of the fabrications are reported in the one-year data and half in the five-year. Assume all of the one-year respondents are braggarts and subtract them plus an additional equal number of presumed braggarts from the five-year positives you have a fair number of presumably honest respondents left. What sort of DGU computation do you get then?

About half a million.

be equally likely to fall into any of the five years and you would expect about 20% to occur in the past year. In fact Kleck found 40% said that it had occured in the past year, which suggests that there is more fabricating than truth-telling going on. Because 40% is twice 20%, you get a factor of two difference in the estimate depending on

Perhaps we use the word “fabricating” differently. Honest telescoping would not be fabricating to me. The incident occurred but is misreported as to its exact time. Fabrication suggests knowing manufacture of a phoney event. Can you really be sure that the 40% one-year figure strongly supports your conclusion.

I am assuming that both forgetfulness and telescoping operate to provide a net overall reporting error that distorts not only the overall total of reported incidents but which is different at different points in the survey period. Suppose 100 actual events each year. Honest telescoping leads respondents to overstate actuals in thet first year by 20%. Each year (before the first) 10% of events are honestly forgotten or supply the “pool” out of which telescoped events get moved forward in time. This would give 2nd through 5th year recollections of 90, 80, 70 and 60 events. This would give 420 honestly if inaccurately reported events in the five year period or 85 per annum while the one year reply rate would give 120, almost 50% higher.

But not 100% higher, which is what Kleck found.

Are such magnitudes of telescoping and forgetfulness unheard of? That’s a real question, not a rhetorical one.

I don’t know.

Does he discount telescoping completely or is my understanding (as exemplified in my 100 event example) wrong altogether?

He believes that telescoping and recall failure tend to cancel out, so the 2.5M estimate does not include any correction for telescoping.

I am assuming that net telescoping would be more likely to be pronounced in the first year and that forgetfulness increases the farther back one tries to remember. Later, you mention that Kleck “cites some data that suggests that telescoping would make no more than a 20% difference” but I don’t know exactly how the difference is defined.

A 20% difference in the estimate.

In trying to resolve the discrepancies between Kleck and the NCVS regarding to shoot) whereas Kleck’s respondents include not only NCVS types but, say, the female apartment dweller who refers to her handgun to send a persistent stranger away from her door.

Kleck take pains to only count DGUs against actual crimes.

Did Kleck’s interviewers go into the same excruciating detail as the NCVS interviewers? The NCVS intentionally tries to elicit false positives by asking respondents to report incidents even if they weren’t sure a crime had been committed and then tries to weed these out. I am sure Kleck tried to screen out nervous people going to investigate tree branches scraping against windows but I had not thought his survey was as detailed as the NCVS regarding the crime victimization.

Not as detailed, but he only counted those cases where the respondent identified a particular crime that was being attempted.

The top of the NCVS range for B seems rather high, but 30% seems like reasonable number for B. Kleck’s survey gives numbers for B that are way too low.

But the reasonableness of the 30% figure lies (only?) in the fact that it is between the police 37% and the criminal 18%, yes? And the unreasonableness of Kleck’s 2% to 5% derives (only?) from the fact that it is so much lower than the criminals’ 18%, yes? Actually, it derives partly

because he felt forgetfulness was more of a problem than telescoping. Does the final published version note the lower five-year estimate?

Afraid not. The lowest estimate he has is 1.2M, and it’s clear that he regards the 2.5M one as best.

Well, I’ll talk to him about that. (The 1.2 million is his five-year estimate or a lowest one-year limit?)

Five year estimate.

In any event, your calculation should prompt us to wonder how civilian DGUs ought to compare with police and criminal “success” rates. If the

For example, a police officer is under a duty to stay engaged with a

Civilian DGUs ought to involve fewer shots, on average.

Not necessarily… The attacker is probably more motivated to attack his victim than to fight third parties (like police).

(This would suggest criminal attacks are unusually deadly in intent.

Only as compared to criminal attacks on police when they intervene.

While I’m here, this is my estimate of the number of DGU-related woundings.

The NEISS estimates 60k Assault/legal intervention nonfatal firearm injuries treated in hospitals. According to the NCVS about 90% of those wounded in this way get hospital treatment, so there are roughly 70k such woundings each year. Add in 18k deaths and I estimate about 90k assault/legal intervention shootings altogether. Using the data from Table 4.2 (on Civilian Legal Defensive Homicides) as Kleck does, about 10-13% of the woundings will be self-defence, or 9k-12k, i.e roughly 10,000.

If I read NCJ-147003 correctly it states that in 1992 offenders armed with handguns committed a 931,000 violent crimes I assume NCVS, not UCR, figure). The report gives the following shooting/hit rates based on 1987-1992:

 Shot at victim         16.6%
 Hit victim              3.0
 Missed victim          13.6

Three per cent of 931,000 is 27,930. If 90% of these incidents send the victim to the hospital that accounts for 25,137 nonfatal firearms injuries treated in hospitals. If another 10,000 are civilian legal defensive woundings (which assumes all injured criminals go to the hospital–a matter in dispute) we account for only half of your 70,000 figure: who is taking up the other hospital beds?

70,000 is my estimate of the total woundings. The number treated in hospital is 60,000.

The NCVS is known to undercount gunshot wounds by a factor of two. Cook believes that the discrepency is caused because many of these victims are criminals who would not agree to participate in NCVS. This is why I believe that the NCVS could undercount DGUs by as much as a factor of two. As well as the 80,000 DGUs that the NCVS detects, there could be another 80,000 by criminals defending against other criminals.