4523 Auhay Drive
Santa Barbara CA 93110
[This document, identified by title or as NCJ 185056, is in the archive/of the National Criminal Justice Reference Service. Write to NCJRS, Box 6000, Rockville MD 20849-6000 or call 1-800-851-3420. The abstract is available at www.ncjrs.org.]
For the entire data set the OGU rate was 31.1% for men, 9.6% for women. There is considerable variation from year to year in the range 27-34% for men, 6-13% for women. Both series show a general upward trend, here summarized with simple linear regressions of % Yes on Year. For men, the average annual increase in this percentage is 0.105, implying a positive change of 2.2 over the 21-year period (32.2% in 1994, up from 30.0% in 1973). The same trend is more pronounced for women, 0.223 per annum, or 4.7 over the 21-year period (12.0% in 1994, up from 7.3% in 1973).
The population of veterans has remained relatively constant in absolute numbers over the period considered here. Decennial census reports show 28.1 million veterans in 1970, 27.4 million in 1980. The 1992 National Survey of Veterans (U.S. Department of Veterans Affairs 1995) estimated the veteran population as 27.4 million, 7.8 million of whom had served in the United States only. Of the 19.6 million serving outside the country, an estimated 48.2%. or 9.45 million, would have answered "Yes" to the question, "During your military service were you ever in or exposed to combat?" Hence, an estimated 11.1% of the 85.2 million men 20 years old and over in 1992 (U.S. Bureau of Justice Statistics 1994b) had experienced combat conditions. The GSS regression estimate of 32.2% of men with OGU experience as of 1994 is reduced to 21.1% if those with combat experience are excluded. This may be an excessive correction. "Exposed to combat" is a broader category than "threatened with a gun or shot at." Moreover, the correction unrealistically assumes that no OGUs apart from those in the combat setting were experienced by these veterans. In any event, combining the corrected rate for men with the regression estimate for women implies an overall GSS OGU rate of 16.4% in 1994, weighting the sex-specific rates by U.S population estimates (U.S. Bureau of Justice Statistics 1997).
The subsequent exploratory analysis of OGU rates in GSS exploits only two classifications, in addition to sex of respondent, that were used in designing a set of tables: birth years are grouped into 5-year intervals. Responses are described with the aid of the question immediately following the gun threat question: "Did this happen to you as a child or as an adult?" The categories, therefore, are No (to initial question); Yes, as a child; Yes, as an adult; and Yes, both. Two rates are defined, % Ch, with Child plus Both as the numerator, and % Ad, with Adult plus Both as the numerator. There is only a small overlap of these two variables, which will be analyzed separately.
Table I pertains to just one of the 14 birth cohorts considered here, those born in 1950-54. This cohort is recent enough to have had relatively little involvement in military service. As of 1990, 14.9% of the men in this cohort were veterans who served during wartime, all of them in the Vietnam era (U.S. Department of Veterans Affairs 1991 and decennial census age distribution). Whatever the contribution of this experience to the OGU count in GSS, we are hereafter mainly concerned not with the lifetime OGU rates as such but with changes therein within cohorts over time. If veterans consistently reported combat experiences as OGUs, that bias would wash out in a comparison between years.
Fortuitously, 1950-54 is also the cohort with the largest sample counts, 1,007 men and 1,215 women. At the time of their first reports on OGU, persons in this cohort were on average about 21 years old, so their experience as adults was meager. But already 6.9% of the women and 23.4% of the men had ever been OGU victims. After 21 years, these percentages had risen to 17.5% and 34.9% respectively. The annual increments were 0.51 for women and 0.55 for men. The 1973 and 1994 rates are estimated from the regression lines shown in Figure 1. To take account of the variations in sample size among the Surveys, these regressions were calculated with Year as the independent variable; the binary dependent variable is Y = 1 if the respondent reported OGU as an adult or as both an adult and a child, Y = 0 if not.
It should be emphasized that in GSS the samples are drawn independently in each survey year. There is no reinterview procedure or panel design as in NCVS. Hence the sample responding in one year will not overlap with the sample of respondents in another year, except for the rarest kind of random accident. With samples as small as 25 and no larger than 124, we can expect wide variation in sample OGU rates from year to year, and that is confirmed by the scattering of data points in the figure. The regression calculation irons out these random fluctuations to provide a sense of the change that occurred in the entire population of this cohort—some 20 million strong, as of 1990—during the 21-year period 1973-94.
When this technique is applied to all the cohorts (except those born before 1905 or after 1974), the regression estimates shown in Table II are obtained. The initial year is 1973 for the cohorts born in 1950-54 or earlier, and 1978, 1983, 1988, or 1993 for the subsequent cohorts. It is helpful to distinguish three groups of cohorts. The estimates for the five earliest cohorts are erratic. Of the ten regression coefficients, five have anomalous negative signs. We should keep in mind that the people in these five cohorts were in their mid-40s to mid-60s in 1973 and their mid-60s to mid-80s by 1994. In all likelihood a few of them encountered OGUs after 1973. But some of these will have been second or third such experiences; only the first OGU would increment the count of those having "ever" experienced gun threat. The safe conclusion is that the data are not sufficient to estimate at all reliably what was most likely a slight increase during the two decades. In other words, "No comment" as to what happened in these early cohorts. For the remaining cohorts, all the coefficients are positive and, as a general tendency, the more recent the cohort, the greater the annual change, as would be expected if first experiences with an OGU incident are more probable among younger persons. However, the male cohorts born in 1930-34 and 1940-49 have proportions of wartime veterans ranging from 29% to 47%. This should not appreciably bias the slopes of the lines, which are largely determined by OGU experience following the end of the Vietnam War; but both the 1973 and 1994 estimates may be upwardly biased as compared to rates that would have been obtained with a question ruling out military experience. Finally, in the five most recent cohorts the veteran proportions are below 15%. As it happens, it is these same cohorts that provide readings of OGU rates as early as age 21.
As of 1994, there is a general pattern in intercohort comparisons for both men and women. The lowest rates are observed for the earliest cohorts. The rates rise to a peak for cohort 1945-49 for men and 1950-54 for women and fall off thereafter. This result needs to be interpreted cautiously, as far as the male data are concerned, for the reason already indicated. For women, however, the implication is clear: there is a rising trend of OGU rates over a period of some four decades. Moreover, there is no strong reason to think it had spent itself as of 1994, inasmuch as the four most recent cohorts were then at ages where additional OGU experience was to be expected.
The estimates of annual change in Table II are, in terms of dimensions, strictly comparable to NCVS or any other data recording annual victimization rates for the preceding year: number of events per capita per annum. (For references on dimensional analysis See Duncan 1984, pp. 159-62.) Of course, a negative change, unless it is produced by some turnover in the population during the period between observations, is impossible. But when the successive observations are made on independent, small samples—and especially for cohorts at advanced ages where the population changes are likely to be slight—an ostensible decline in the lifetime OGU rate is not at all unlikely. When the 28 estimated changes in lifetime OGU rates are applied to the total population distribution by age and sex, the result is not unreasonable: 918,000 OGUs in a year. For comparison, NCVS estimated 1.3 Million gun crime victimizations in 1993 (U.S. Bureau of Justice Statistics 1995), about three-fourths of which, or 975,000, involved respondents 20 years old and older; that fraction is based on 1987-92 age-specific data pertaining to handgun crime victimizations (U.S. Bureau of Justice Statistics 1994a).
I make only two minimal claims concerning this comparison. First, the calculation using GSS data has a coherent demographic rationale, albeit the assumption made in applying it—that random sampling error is the only or overwhelmingly most important source of distracting variability—is vulnerable. There are, of course, the usual questions that arise in cohort analysis about cohort closure and differential survivorship as a cohort ages (for references and remarks on cohort analysis see Duncan and Stenbeck 1988); but there is no specific information concerning the example at hand to add to previous discussions of these issues. There are also the well-known cautions concerning the fallibility of memory and response biases that plague all surveys, not excluding five surveys providing the estimates of proportions of persons ever involved in DGUs that were "adjusted" by Kleck and Gertz (1995, Table 1) so as to produce "estimated annual number of defensive uses of guns ... for U.S., 1993." One would have expected the authors to heed their own caution (p. 157) that questions with a "lifetime recall period" make it "impossible to estimate uses within any specified time span." Instead, they accomplished the impossible by concatenating several "adjustments" (described in Kates and Kleck 1997, pp. 187-90) that range from the arbitrary to the absurd. To illustrate: Using figures from the 1975 Field survey Kleck computes the ratio of 1.4% with DGU, one-year period, to 8.6% with DGU, lifetime recall period, to obtain Adjustment B = 0.163. He applies it to the lifetime DGU rates in the aforementioned five surveys to convert them to a one-year basis. The same procedure, using the same source (see Table III, below) to compute an adjustment factor for OGUs, yields the ratio 2.8/13.0 = 0.215. Hence the GSS lifetime OGU rate of 16.4% (roughly estimated combat OGUs deleted) implies an annual OGU rate of 3.5%. The implied count of OGUs in 1993 is 6.4 million.
My second claim concerns the relevance of the comparison of GSS (918,000) and NCVS (975,000) frequencies of OGUs. It would have been different if, for example, I had cited NCS data for a date before the NCVS instituted the redesign, which generated higher levels of violent crime than were reported earlier (U.S. Bureau of Justice Statistics 1997b). I only assert, therefore, that the GSS annual OGU rate inferred from the cohort analysis is in the same ball-park as the NCS/NCVS rates. To that extent, the two surveys provide mutual support for each other's findings. The data cannot be forced into making a decision that one or the other over- or underestimates annual OGUs. But the extravagant claim about NCVS "exaggeration" (see section 3.1, below) is certainly called into question. Nor do I suggest that the GSS question is a useful technique for measuring period-specific OGU frequency. Quite to the contrary—for obvious statistical reasons. But if the GSS question works even passably well in a situation it is, for statistical reasons, poorly suited to deal with, it works much better in measuring what it is intended to measure. Whether lifetime OGU rates are worth estimating is a substantive question. For some purposes they are not; for others, they may be.
Turning to the data on OGU encounters in childhood, we have a straightforward display of quantitative oral history in Figure 2, with no concern about military experiences and with cohorts comparable as to age at the time the incidents were likely to have occurred. (For sample sizes, see Table II.) Over a 65-year period, from about the time of World War I (when the earliest cohort was in the later years of childhood) to the 1980s, the proportion of persons who completed their childhood having experienced a gun threat or having been shot at rose from 1.5% to 11.0% for boys and from zero to 3.4% for girls. (For the earliest cohort the proportion of women encountering OGU as a child was literally nil in the sample and a negative -0.2% according to the regression estimate.) Whether these trends continued into the 1990s cannot be investigated with these data.
For comparison with the findings just reported, the most useful sources are two special reports (U.S. Bureau of Justice Statistics 1990, 1994a) showing handgun victimization rates computed from data collected in NCS, 1979-87 and 1987-92, respectively. I am using the tables showing separately for male and female victims the annual average handgun victimization rates at ages 12-15 and 16-19. (To approximate the rates for all males and all females, not shown explicitly in the table for 1979-87, I averaged the rates for white and black victims, using as weights the implied age-specific proportions white and black in the 1987-92 table.)
Inasmuch as childhood is not defined for GSS respondents, I shall assume that it ends at the eighteenth birthday. To simulate cumulative handgun crime rates with the NCS data, I consider a synthetic cohort that is subjected to the annual rate for persons 12-15 years of age for four years and to the rate for persons 16-19 years old for two years. The 1987-92 NCS data imply lifetime victimization rates to age 18 of 4.8% for males and 2.0% for females. The regression estimates of childhood OGU rates in GSS are 11.0% for males and 3.4% for females in the cohort born in 1970-74. The 1979-87 NCS simulated rates are 3.1% for males and 1.4% for females, as compared to 9.9% for males and 3.0% for females, averaging the GSS regression estimates for cohorts born in 1960-64 and 1965-69.
There are at least three known biases that could materially affect these comparisons. The NCS data include only handgun victimizations, which amount to perhaps 80% of all gun victimizations. They exclude victimizations occurring prior to age 12, the numbers of which can only be guessed at. To the opposite effect, GSS excludes victimizations subsequent to the first one; but nothing is known about repeat victimizations in childhood. Methodological factors, possibly significant in terms of the size of their effects but uncertain as to direction, should not be overlooked. NCS/NCVS interviews each person age 12 or older in every sampled household. GSS interviews a randomly chosen adult in each household. And the GSS question pertains to experience not in the year immediately past but during a period more or less remote in time from the interview. My hunch is that adolescents, particularly when interrogated while family members may be present, may neglect to report incidents that in later years they are willing to recall.
In any event, the intercohort trends in childhood OGU rates disclosed by the GSS data are challenging. It would be a mistake to write them off simply because they are higher than the rates derived from NCS data. (It will be interesting to learn what effect the NCVS redesign had on reports of gun crime victimizations at the younger ages, whenever the Bureau of Justice Statistics reports the 1993 and subsequent frequencies of gun crimes by age of respondent.)
Kleck and Gertz (1995, Table 1) present data from three surveys in which the screen question is worded in much the same way as in NSDS. The question pertains to DGUs occurring in the past five years to either the respondent or to another member of her or his household. In the Hart and Mauser polls, as in NSDS, the initial screen question was followed by a question on whether the DGU was to protect against an animal or a person. In the Tarrance poll, the initial question specified protection against a person. Whereas Hart specified use of a handgun, the others referred to "a gun" without qualification.
In the 1993 NSDS (n = 4,977 weighted observations), the initial question produced a DGU rate of 4.9% of households, prior to any exclusions. With animal defenses excluded, the rate was 4.5%. Restricting the count to interviews with fairly complete information" yielded a rate of 4.3%. Further restrictions, not clearly described, led to the Series A rate of 3.9%. Restriction to positive responses "devoid of any problematic indications" produced the Series B rate of 3.5%. For comparison the Hart poll in 1981 (n = 1,228) obtained a rate of 4% (shown only to a single digit) using a handgun. Kleck's adjustment to simulate the rate for uses of all guns yields an implied rate of 5%. Both of these figures are subject to an uncertainty of ±0.5% simply because of rounding error in computing the sample proportion, quite apart from sampling variability or other sources of error. The Mauser survey in 1990 (n = 343) found a rate of 3.79%, and the 1994 Tarrance poll (n = 1,000) obtained a rate of only 2% (again subject to a relatively large rounding error).
The relatively high screen question rate of 4.5% in NSDS, as compared to the other three percentages of 5%, 3.79%, and 2%, cannot be shown to follow from some technical superiority of NSDS. The other surveys are not sufficiently well documented to permit any inference about technical quality. The NSDS does, of course, set a new standard for documenting some uncertainties attributable to responses of substandard quality. But these emerge in the questions following the initial screen. The screen question (Q 4. in the interview) comes out of the blue, just as it must have in the polls by Hart, Mauser, and Tarrance. By the same token, the latter must be assumed to be subject to the same sorts of data quality problems that are brought to the surface in NSDS. Their estimated rates are therefore overstated relative to those they would have obtained with NSDS-like incident questions.
By the same analogical reasoning, one surmises that all the OGU rates, except NCVS, cited herein would be subject to downward revision if the same questions were followed by detailed inquiries about the nature and circumstances of the OGU incidents. But as long as we compare OGU with DGU rates both derived from screen questions, it is legitimate to assume that both are subject to upward biases, in this sense, though not necessarily to the same degree.
Unfortunately, the results obtained with a 5-year period for household DGU incidents have no parallel in the OGU domain. Hence the NSPOF results—where the DGU screen question pertains to persons rather than households and to a one-year period—may be somewhat more relevant. The statistics are presented in generous detail by Cook and Ludwig in their Technical Report (1996, Table 6.1) and more sketchily in their 1998 article. The 1-year DGU rate from the screen question was 1.93%, dropping to 1.64% with exclusion of defenses against animals. Further exclusions, along the lines of the criteria enumerated by Kleck and Gertz (1995, p. 162), produce rates of 1.44%, 1.29%, 0.95%, and 0.77%. Finally, dropping one sample case of an incident related to work yields a DGU rate of 0.75%. The authors imply that 0.77% is the estimate most nearly comparable to Kleck's 1-year Series A rate for persons, 1.326%. Kleck (1997, Table 5.1), however, in comparing NSPOF and NSDS, cites the 1.44% for the former, 1.326% for the latter. The spectator in this contest of opinions might prefer to compare the NSPOF 0.77% with the NSDS.Series B rate of 1.125%. The two parties seem to agree that the differences between the two surveys are within the range of probable sampling error. But that is not the issue here. Rather, the disagreement about the reading of the NSPOF estimates suggests that comparability may be uncertain even when nominally similar criteria for classifying incidents are employed. Whatever the nominal criteria, the principal investigator in all these surveys works with the coded data generated by an interviewing process the control and standardization of which are always problematic and often undocumented, although NCVS goes about as far as one can well imagine in implementing safeguards against interviewer bias and variability.
The California (Field Institute) data consistently show OGUs more frequent than DGUs, for both men and women and for reporting periods of one year, two years, or lifetime to date. The excess of OGUs over DGUs, however, is much greater for men. Or, to describe the same interaction another way, the sex effect on OGU rates is larger than on DGU rates. This data set is the only one I know that provides such a comparison. Kleck and Gertz (1995, p. 178) report as their "most surprising finding" that 46% of the DGUs in their sample were reported by women. This finding need not have surprised them nor have been described as "questionable" by Smith (1997, p. 1467). Within errors of rounding, the California data imply the same percentage for the one-year period, albeit lower percentages for the two longer periods. The substantial sex effect on OGU rates found in the California data is consistent with GSS data already cited and with NCS data showing gun crime victimization rates for men two or more times as high as those for women (U.S. Bureau of Justice Statistics 1986, 1990, 1994a), although the rates for both men and women are much higher in the California survey than in NCS.
One other organization has provided comparisons of rates based on life- time experience. In the May 1991 Gallup Poll, the 211 "pistol owners" identified in the sample of 1,002 adults were asked, "Have you ever used your handgun to defend yourself, your home, or family or possessions, either by firing it or threatening to fire it?" and "Has a handgun ever been used to threaten you in a robbery, mugging or some other situation?" The DGU rate was 8% "Yes" to the first question, and the OGU rate was 16% "Yes" to the second (Hueber 1991). Neither question explicitly excludes a military context, so they are comparable in that respect. In December 1993 the same organization reported comparisons for a special sample of gun owners—adults "who say there is a gun in their home or on their property" (Unsigned 1993). Apparently this sample (hereafter 1993b) overlapped the cross-section of 1,014 adults interviewed at the same time (hereafter 1993a). The relevant questions were: Q 54 "Not including military combat, have you ever used a gun to defend yourself, your home, your family or possessions, either by firing it or threatening to fire it?" and Q 56 "Not including military combat, has a gun ever been used to threaten you in a robbery, mugging or some other situation?" The OGU and DGU rates, respectively, were 18% and 15% for all 841 owners, 19% and 20% for all 533 handgun owners, and 15% and 7% for the 269 owners of long guns only. The discrepancy in regard to sample sizes is not explained. Nor is it obvious why the OGU/DGU comparison is so different in the 1991 and 1993b samples of handgun owners. The 1991 and 1993b OGU rates are similar, but the DGU rate in 1993b is much higher than in 1991. In any event, neither survey shows DGUs significantly more frequent than OGUs.
As for comparisons based on incidents occurring in the past year, Table III includes this information from the 1975 Field survey. NCVS, though it was not designed for such a purpose, has yielded OGU and DGU counts as byproducts of its inquiries about victimization. As of 1993 the OGU rate was 0.48 gun use incidents per 100 persons age 12 and over (U.S. Bureau of Justice Statistics 1996). The comparable DGU rate—based on average annual DGUs, 1992-94, as given by Cook et al. (1997, p. 468)—was 0.051.
According to an unpublished tabulation by Gary Kleck of data on responses to two questions in the 1994 NSPOF survey, 141 of the 2,568 respondents—a victimization rate of 5.5%—answered "Yes" to Q 51: "During the past 12 months has anyone robbed, tried to attack, or attacked you?" When these victims were asked Q. 52, "Did the perpetrator use a gun in any of these incidents?" 9 of the 141, or 6.4%, answered "Yes" and 7, or 5.0%, "Don't know." These frequencies can be compared to the 1994 NCVS incidence rate of 4.58% for crimes of violence, with 10.9% of these crimes classified as "Weapon used" and 8.9% as "Don't know if weapon present" (U.S. Bureau of Justice Statistics 1997a). The NSPOF violent crime incidence rate is in reasonable agreement with the NCVS rate, but the percentage of gun use is lower. The OGU rate in NSPOF is 9/2,568 = 0.35%, a bit lower than the NCVS 0.48%. But the NSPOF DGU rate, variously estimated as 0.77% to 1.64% (depending on which criteria are met) is, of course, much higher than the NCVS 0.051%. NSPOF is, therefore, unique in finding a large excess of DGU over OGU. There is an obvious but not necessarily complete explanation. NSPOF is the only OGU survey in which the question put to respondents does not include "threatened with a gun," "threatened by a person with a gun," or similar explicit reference to threat. In light of NCVS data indicating that a large fraction of gun crimes involve only threats (59.4% in 1994), it seems that the NSPOF question significantly narrows the scope of the category. In support of this suggestion, I note that of the 8 NSPOF defenders facing gun-armed offenders (in the data obtained under the most restrictive definition of DGU), 4 responded "Threaten only" to Q 75, "Did the perpetrator threaten, attack or injure you?" Another 2 responded "Attacked but did not injure" or "Attacked and injured," and 2 reported neither threatened nor attacked (Cook and Ludwig 1996, Table 6.10a.b). Something more could perhaps be learned from a cross-classification of Qs 52, "Did the perpetrator use a gun in any of these incidents?" (i.e., the victimizations in the last 12 months. Q 51) and Qs 78-79, "Did the perpetrators) have any kind of weapon?...What was the weapon?" pertaining only to incidents in which the respondent defended with a gun. It is odd that the 19 gun defenders reported as many as 8 gun-armed offenders while the entire sample of 2,568 respondents reported only 9 violent crimes by such offenders. Pending more detailed analysis going beyond the published data—or alternative explanations of the anomalous OGU rate in NSPOF—I propose that it be ignored.
In the 1975 survey Field found a 2:1 ratio of handgun OGUs to handgun DGUs in the past year. The 1979 and 1981 Field OGU rate of 3% (all guns) is over twice as high as the NSDS DGU rate of 1.326%. But the latter could be raised by a factor of 4.5/3.9 = 1.15 to simulate the rate that would have been obtained with a single screen question for persons (see section 2.2). Such an adjustment leaves the ratio of Field OGU rate to NSDS DGU rate at about 2:1.
It may be well to include in the record the information we have on the frequency with which DGUsers encounter gun-armed offenders. In NSDS 18% of c. 202 defenders recalling their most recent incidents answered "Yes" and specified handgun or other gun (possibly both) when asked, "Did the offender(s) have a weapon....?" (Kleck and Gertz 1995, p. 175). Similarly, the NSPOF tabulations given by Cook and Ludwig (1996, Table 6.6, derived from Qs 78-79) imply that about 20% of DGUsers in the last five years faced a perpetrator who had a gun. (The figure is approximate because of inconsistencies between percentages based on weighted estimates and raw sample counts, so I conjecture.) In the drastically pruned sample of last-year DGUsers, 8 of 19, or 42%, of the defenders faced gun-armed offenders. NCVS data for 1987-92 show an OGU rate of 26.7% among victims using firearms to defend themselves or their property—that is, as in NSDS and NSPOF, they "face an offender who also had a firearm," although there is no specification of the way in which (s)he may have used it. The handgun victimization rate (homicides excluded) for the whole population aged 12 and over in 1992 was 0.44%. (U.S. Bureau of Justice Statistics 1994a). DGUsers are a specialized population in regard to victimization by gun-armed offenders. They attract—are attracted to?—OGUsers in highly disproportionate numbers. Kleck and Gertz (1995, p. 175) note that in "only 3%" of DGU incidents both parties shoot. But that is one-sixth of the incidents in which two armed parties face each other, or about 76.000 shoot-outs, enough to account for a large fraction of the 18,000 gun homicides in 1993.
Turning to comparisons derived from questions about lifetime experience, there are enough readings to let the game of selective comparisons go on a long time. There is heterogeneity among surveys in regard to both OGU and DGU rates and also in regard to wordings of questions and other specifications of the survey design. I shall simply list the surveys in roughly chronological order, citing the OGU or DGU rate obtained in each, the source, wording of the question, sample size, and any significant restrictions, as to population or circumstances covered.
Just inspecting the arrays of rates—a half-dozen each of OGU and DGU—one is not inspired to make any strong pronouncement as to which category of use is more frequent in the recent experience of the American population. If required to choose the "best" figure in each set, I would opt for the 16.4% OGU rate from GSS and the 13.6% DGU rate from NSPOF. Neither screen question is confined to a special population or circumstance of the use; but in NSPOF the DGUs may include some defenses against animals, and military combat DGUs are not excluded.
The remainder of Kleck's argument for NCVS exaggeration reduces to two criticisms. One is that NCVS respondents may be mistaken in reporting that the offender(s) possessed a gun. The other is well described in Kleck's own phrasing as a "definitional problem."
Concerning respondents' errors, Kleck notes some possible reasons for suspecting false positives. No doubt some respondents do guess about the offender's weapon—some guess right, some wrong. But Kleck ignores the possibility of false negatives and bypasses the question as to which may be the more frequent error, notwithstanding his own condemnation of "one-sided speculation" in such a situation (Kleck and Gertz 1997). Kleck does not quantify his speculations, but there is quantitative evidence that the redesigned NCVS provides ample opportunity for respondents to acknowledge their own uncertainty in reports of gun possession. In the 1993 annual report the published percentages imply an estimate of 1.02 million gun crime incidents altogether, of which 9,900 were "gun type unknown" (U.S. Bureau of Justice Statistics 1996, Table 66). There are an additional 119,000 incidents with "weapon used, weapon type unknown." The latter could have been guns. Furthermore, in 683,000 incidents the response was "Don't know if weapon present"; 514,000 of these were classified as simple assaults, hence subject to reclassification as aggravated assaults if it were determined that a weapon was present. In regard to the possibility that significant numbers of the "Don't knows" pertained to offenders who were actually carrying guns, consider this finding in NSPOF. All the defenders had, by definition, made some "use" of their guns, if only to mention them to the offenders. Yet only 55.2% said "Yes" to Q 63, "Did the person (people) you were defending against know you had a gun?" (Cook and Ludwig 1996, Table 6.6).
There is, then, a large reservoir of potential false negatives which are offset by nothing except Kleck's guesses about the respondents' erroneous guesses. Let us recall his maxim, "An ounce of evidence outweighs a ton of maybes" (quoted in Dizard et al., 1999, p. 334). Kleck has not provided the evidence needed to conclude that the NCVS estimate of 1.02 million violent crimes committed by offenders possessing guns is a serious exaggeration as far as the count of incidents involving gun possession is concerned.
To the definitional issue: It should be stressed that NCVS determines victimization in a violent crime by ascertaining whether the respondent was attacked, was the subject of an attempted attack, or was threatened with harm, irrespective of whether a weapon was involved. The presence of a weapon does not ipso facto affect the determination that a crime occurred; but it does bear upon the classification of assaults. An aggravated assault, as distinguished from a simple assault, is one in which there is a serious injury and/or a "threat or attempted attack by an offender armed with a gun, knife, or other object used as a weapon, not resulting in victim injury" (U.S. Bureau of Justice Statistics 1996, "Glossary," p. 172, emphasis added). Kleck, however, proposes a distinction between incidents in which "a threat with a gun actually occurred, as opposed to the offender merely possessing a gun during an incident in which he made a threat not involving a gun" (Kates and Kleck 1997, p. 174). The distinction seems to rest on a fine point. "Give me the money or I'll kill you," as I understand the argument, is "a threat not involving a gun," even if the offender has a firearm, whereas Give me the money or I'll shoot you is a threat with a gun." Only the latter would be a gun crime if Kleck's distinction were accepted. It would be a nice problem for the designer of a survey to come up with questions enabling respondents to distinguish reliably between threats of gun-armed offenders and "threats with a gun." Whatever the legal or other relevance of Kleck's distinction, expecting a respondent to comprehend it—much less apply it accurately in recalling a confusing, stressful experience from some months in the past—is unrealistic.
I am not forgetting that in NSDS defenders were asked to distinguish between several kinds of actions they may have taken with their weapons. But we have no evidence on the reliability with which they accomplished this task. Kleck himself questions the accuracy of their reports about wounding the offender; nor does he explain how a "warning shot" differs from a shot that happened to miss the offender. Moreover, a double standard is applied in his discussion of this issue. He objects that NCVS does not inquire whether the gun was "even verbally referred to in a threatening manner (for example, "I've got a gun. and I'm going to use it")" (Kates and Kleck 1997, p. 175). But one of the answer categories in the NSDS question on what R did with the gun was "Verbally referred to gun ("I've got a gun," etc.)," with no suggestion that the utterance must specify a threat. Kleck requires the offender in NCVS but not the defender in NSDS to say, "I'm going to use it."
Table 3A in Kleck and Gertz (1995) shows that 75.7% of NSDS defenders "brandished or showed gun," with lesser percentages pointing or firing the gun, or wounding or killing the offender—all of which could well have been preceded by "brandishing." Hence, the data, as presented, do not exclude the possibility that as many as 24.3% only "verbally referred to gun"—without even doing it "in a threatening manner," as just noted. The criterion on which NCVS is deemed to generate an "exaggeration" forces acknowledgement that NSDS estimates of DGUs may be similarly exaggerated for the same reason. Kleck's statement (1977, p. 156) that in NSDS "Rs claiming a DGU nearly all directly confronted their adversaries and, at a minimum, pointed their guns at them or referred to the guns verbally in a threatening manned (emphasis added) is not consistent with the question wording, answer category, and illustrative response shown for Q 8 in the NSDS questionnaire, where there is no mention whatsoever of "threat." The same is true of NSPOF, where Q 64 asked, "What exactly did you do with the gun in this defensive incident?" The first answer category, "Tell the perpetrator(s) you had a gun" (no threat specified), was mentioned by 37.2% of 85 defenders in their most recent incident. As in NSDS, the modal response was "Show the gun to the perpetrator," mentioned by 68.7%. Assuming with some plausibility that "show" preceded all the more serious uses, 31.3% may have "told" without threatening, as far as one can infer from the verbatim text of the questionnaire (Cook and Ludwig 1996, Table 6.6).
To ascertain whether Kleck's objection to the NCVS definition of a gun crime has any quantitative importance, we would need a controlled experiment on question wording, comparing the OGU rates obtained with a question specifying that a gun was used. with an otherwise identical question specifying only that the offender had a gun. Lacking such an experiment, we may review the surveys reporting lifetime OGU rates. Four of them specifying gun use (Gallup 1991, 1993b, Cambridge Reports and GSS) yielded estimates of 16, 18, 11, and 16.4 percent (mean 15.35) The other five surveys asking only about a "person with a gun" or "someone with a gun" (Field 1975, 1979, 1981, and Time/CNN 1989, 1993) obtained rates of 13.0, 10, 13, 17, and 15 percent (mean 13.6). The means would be 17.0 and 14.25 if the rates in the three surveys specifying handgun rather than simply gun were raised by a factor of 1.25, Kleck's Adjustment A (Kates and Kleck 1977, p. 187). The best available evidence—indeed, the only evidence—fails to support Kleck's argument.
The culmination of Kleck's indictment of NCVS is his summary statement, "Up to 83 percent of the incidents labeled violent 'gun crimes' in NCVS publications may not have involved any gun use of any kind" (Kates and Kleck 1997, p. 175). And that is where the matter is allowed to rest as far as this particular publication—and those of its readers who have not seen Kleck's other writings are concerned.
These other publications must be consulted for more statistical detail and qualifications of the argument. Kleck's source for the description of handgun crimes is a 2-page special report showing estimates of victimizations by offenders with handguns for the period 1987-92 (U.S. Bureau of Justice Statistics 1994a). Some 16.6% of the handgun crimes involved the offender shooting at the victim, and these are the only NCVS crimes "that one can be confident involved offenders actually using guns" (Kleck 1997, p. 6). Another 46.8% were classified as "Weapon present or threatened with weapon." These "ambiguous" crimes for which "it is impossible to determine whether the victim was actually threatened with a gun or merely reported that the offender possessed a gun" (p. 6)—are nonetheless "generously" assumed to have "involved guns being used to threaten the victim" (p. 7). The remaining 36.6% of the handgun crimes were originally classified as Other attack (19.9%), Verbal threat of attack (15.4%). Other threat (0.8%), and Unknown action (0.5%). For all of these categories, "there is no indication at all that the gun allegedly possessed by the offender was actually used in any way" (p. 6).
There are some slippages when Kleck applies these percentages to obtain estimates for 1992 (Kleck and Gertz 1995) and then for 1993 (Kleck 1997). The Bureau of Justice Statistics report counted victimizations; Kleck estimates number of incidents. The source considered only handgun crimes; Kleck offers estimates for all gun crimes. The 1987-92 data were collected before the NCVS redesign; but Kleck applies the percentages computed from these data to the 1993 NCVS count obtained with the redesign in effect. Concerning the redesign and its impact on the statistics, see U.S. Bureau of Justice Statistics (1996, Appendix II, and 1997b).
An unpublished tabulation of 1993 NCVS data serves to update the 1987- 92 findings, counting incidents rather than victimizations, and including both handguns and other guns in the counts (see Table IV). There are 627 sample incidents altogether, including 548 handgun crimes (38 of which also involved other guns or other weapons) and 79 crimes involving guns other than handguns, but excluding reports of gun type unknown. These sample counts correspond to the published estimates of 1,010,000 gun crimes (same exclusion) comprising 881,000 handgun crime incidents and 129,000 incidents with other guns (U.S. Bureau of Justice Statistics 1996, Table 66). The distribution by Kleck's three action categories, shown in Table IV, is 10.5% Shot at (both hits and misses); 37.6% Weapon present, the "ambiguous" category, to be included in the "generous" estimate; and 51.9% in the catchall category, recode (2), of other attacks, attempts, and threats, for which there is "no indication at all" of gun use. The "generous" estimate of gun crimes amounts to 10.5% + 37.6% = 48.1% of the NCVS total count of 1,010,000, or 486,000 as the number of gun crimes of which "we can be confident" on Kleck's criteria. The relevant DGU estimates from NSDS are 2.53 million (Series A) and 2.14 million (Series B)—values slightly different from those in Kleck and Gertz (1995, Table 2), where the cited population base is not the one they actually used. Hence, Kleck's summary (1997, p. 160)—"the best available evidence indicates that guns are used about three to five times as often for defensive purposes as for criminal purposes"—should be changed to read "four to five times." Kleck's own reduction of NCVS gun crimes to 646,000 (1997, p. 7) implies ratios of three to four times.
The disaggregated data in Table IV shed some light on the plausibility of Kleck's exclusion of actions in recode category (2) from his estimate of gun crimes. Bear in mind that a gun is reported to be present in all cases. Some 44.9% of the 325 respondents in this category reported/Verbal threat to kill." That's 26.0% of all the gun crimes not involving shooting, or 23.3% of all gun crimes. The offender has a gun and threatens to kill you. That's not a gun crime?
How realistic are these reports? One piece of evidence is at hand. In the 1993 sample, with 23.3% of the gun crimes involving threats to kill. 11.1% of the 377 crimes by knife-carrying offenders and only 4.2% of the 574 with other sharp objects, blunt objects, or other means of delivering death involved such threats. What produces these contrasts if there is no gun use by offenders making the threats? I mention in passing a coincidence with perhaps too many "obvious" explanations. Some 68.9% of the total of 212 incidents involving threat to kill on the part of offenders with any kind of weapon were perpetrated by gun-armed offenders. In 1993, 69.6% of UCR murders and non-negligent manslaughters were effected with guns; 71.5% of all homicides reported in 1993 U.S. vital statistics were likewise produced with guns (Kleck 1977, Tables 1.2 and 7.3).
Concerning the other action categories, attacks with weapons other than guns (recode 2.3) and various kinds of manhandling (2.4), as well as the other sorts of threat and assault, may certainly be enabled or facilitated by gun possession. The gun may, for instance, warn the victim not to flee but to submit to other punishment. It should be remembered that many gun crimes involve more than one offender—38.3% of them in 1979-87 (U.S. Bureau of Justice Statistics 1990). One offender can hold the gun while another does the dirty work.
As for the allegation about the "ambiguous" Weapon present category (recode 3.0), it is without merit. Two separate questions (Qs 22 and 23 in the incident report NCVS-2) considered together establish the fact, as reported by the respondent, that there was a threat of "harm in any way" and that the offender had a gun and was, therefore, in a position to make good the threat. There is no ambiguity here, just the implementation of a definition of gun crime that is not to Kleck's taste.
Two final comments on the issue of NCVS "exaggeration of gun crimes." First, if there were a strong justification of the claim that 83% of gun crimes or, as Kleck says (Kates and Kleck 1997, p. 174), "even half this" are false positives, not canceled out by any such number of false negatives, the matter would be one of failure of the Bureau of Justice Statistics to do its job. A Congressional inquiry would not be out of order, perhaps taking as precedent the work of the Advisory Commission to Study the Consumer Price Index, which reviewed the work of the Bureau of Labor Statistics. (That commission, appointed in 1995 pursuant to action by the Senate Finance Committee, found that the current inflation rate was overstated by something like one-third.)
Another implication of Kleck's assertion is that conclusions he has endorsed (1997, pp. 225-226) concerning the superior efficacy of OGUs as compared to other ways criminals deal with victims, insofar as they were based on NCVS data, should be revised. The "simple percentage analysis" he mentions must refer to something like Table 5.6 in Kleck (1991), which shows an attack rate for gun-armed criminals of 36.6% as compared, for example, with 52.1% for unarmed criminals. But if the only gun-using criminals are the 16.6% who fired, their attack rate is by definition 100% and the rate for the non-shooters, erroneously thought to possess or be using guns, must be 24.0%, that is to say, lower than that for wielders of knives (42.7%) or any of the other weapon categories (56.1%-62.8%), as well as the unarmed. Kleck's more radical correction of NCVS would imply that the most effective OGU is one in which the victim only imagines that the offender has a gun. Or, to state the matter in another way, the evidence cited by Kleck as demonstrating the effectiveness of gun use by criminals supports the assumption that those reported to be wielding guns really do so.
A striking omission from Kleck's discussion of ratios is any mention of the number of DGUses, which must have exceeded the number of DGUsers, the figure he regularly cites. That is perhaps just as well, because the only information we have about the uses concerns the "most recent" use, and there are no more of them than there are users. Descriptive statistics on all uses—concerning, for example, type of crime or offender's actions—are likely to differ markedly from those restricted to most recent use. Extrapolation from the one to the other cannot be justified by any data now available. In any event, the NSDS ratio of uses to users was about 1.5; this pertains to the most recent incident during a five-year period. The NSPOF estimate preferred by Cook and Ludwig (1996, Table 6.2) is 4.7 million uses by 1.46 million users, or 3.2 uses per user in the preceding year. If it were really important to get the "best" estimate of the DGU:OGU ratio, the DGU count should surely be adjusted upward, just as the OGU count should refer to victimizations, not incidents. In 1993 the ratio of victimizations to gun-crime incidents was 1.3, compared to 1.1 for crimes of rape, robbery, and aggravated assault committed by offenders without guns (U.S. Bureau of Justice Statistics 1995, 1996).
But does it really matter? It seems that whether the DGU:OGU ratio is 1:1, 3:1, or 5:1 is immaterial for Kleck as far as gun control policy is concerned. The 1997 chapter on policy is largely a verbatim recapitulation of its 1991 counterpart. The only significant change in wording in this chapter even tangentially related to the ratios is in the statement that "the use of guns by crime victims to defend themselves is [commonplace and] effective both in preventing completion of the crime and in preventing injury to the victim," (Kleck 1997, p. 383), where the words in brackets do not appear in the 1991 text (p. 429). Not only does the elevation of the DGU:OGU ratio leave the earlier policy recommendations intact. They are not affected either by the widely advertised finding of Lott (1998), that the most cost-efficient means of reducing crime is the passage of right-to-carry concealed handgun laws. Kleck (1997, p. 372) neatly fingers the fallacy in that conclusion.
Perhaps it would make a difference in policy recommendations if the DGU:OGU ratio were sensibly less than 1:1, say about 1:10, as would be the case if NCVS counts of both DGU and OGU were used. In view of the ample discussion by Kleck and others of the possibility that NCVS underestimates DGUs, I bypass that issue and only ask: What if NCVS estimates of OGUs were actually understated, perhaps considerably so, as is suggested by the California data reviewed in sections 2.3 and 2.4? Another piece of evidence to that effect is immediately available in NSDS itself. Some 2.53 million defenders confronted offenders, 18% of whom were estimated to have been armed with handguns or other guns (Kleck and Gertz 1995, p. 175). That amounts to 455,000 confrontations between two armed parties. Very few of them are reported in NCVS, only 22,000 per year in 1987-92 (U.S. Bureau of Justice Statistics 1994a). The allegedly "exaggerated" NCVS count could actually be an undercount to the extent of over 400,000 for this reason alone. (As noted earlier, NSPOF similarly found that about 20% of DGUsers confronted gun-armed offenders.)
About 13% are crimes not recognized in NCVS. One-third are burglaries, for which absolutely no testimony was recorded concerning evidence of breaking and entering. Interviewers put responses into the foregoing crime categories supplied for Q 10, "What sort of crime did you believe was being committed against you when you used the gun defensively? For example, were you being attacked in some way, or robbed, or was someone trying to get into your home?" In the NCVS protocol the type of crime is not decided by the respondent or the interviewer nor is the nomenclature of the crime classification offered to either of them. Instead, some seven questions with over 40 specific answer categories elicit the evidence that is used to create the type-of-crime classification by means of an intricate computer-controlled receding operation. If the respondent says the offender was trying to get in, (s)he is asked Qs 15-16 in NCVS-2, "What was the evidence?" (9 answer categories) and "How did the offender try to get in?" (again 9). It is not consistent of Kleck to require specific testimony as to the presence of a gun in NCVS—"Rs are never asked whether they saw a weapon in the offender's possession" (Kates and Kleck 1997, p. 174)—when he does not even ask NSDS respondents what they saw that led them to "believe" (a word he puts into their mouths) someone was trying to get in.
Indeed, there is an oddity about the NSDS data on burglary. Kleck and Gertz (1995, Table 4) report that 5.5% of all NSDS respondents were burglary victims in the preceding year, and this figure is in reasonable agreement with NCVS data. But 19.3% of the defenders were in this category, while 33.8% of all defenders "believed" that a burglary was being committed in their most recent DGU (Table 3C). Kleck (1977, p. 183) states that "only 12.7 percent of U.S. residential burglaries are against occupied homes." If this proportion applies to 19.3% of defenders, then something like 2% or 3% of all defenders were at home at the time of their burglary victimization. What were the rest of the 33.8% defending against? And how is that figure to be reconciled with the statement by Kleck and Gertz (1997, p 1452) referring to "the 194 reported DGU incidents of which about 40 were linked to burglaries"?
It comes down to this: the type-of-crime classification in NSDS is based on much more casual questioning than NCVS uses even in its screen questionnaire. NSDS simply does not include information about the crime that compares in quality to what NCVS obtains in the incident report. Ironically, though, in ascertaining whether NSDS Rs were victims of gun crimes, Q 12A in NSDS is worded the same as Qs 22-23 in the NCVS incident report (NCVS-2)—the very questions that allow the so-called "ambiguity" in regard to offender's gun use of which Kleck complains. And Qs 78-79 in NSPOF are almost the same.
But NSDS Q 11, "Did the offender [any of the offenders] threaten, attack or injure you?" does not probe the mode of violence as carefully as the corresponding Qs 24-33 in NCVS-2. Hence, there is the possibility that the 46.8% of NSDS defenders who experienced "No threat or attack" (Kleck and Gertz 1995, Table 3E) is a bit higher than would have been reported with NCVS-style questioning. Be that as it may, threat, attack, injure, and getting away with money or property are the only criminal acts about which NSDS respondents were queried. As to the last category, it accounts for 11.0% of "property crimes" (defined for this purpose as burglary, robbery, or theft) or 6.7% of all DGU incidents. Assuming, unrealistically, that all of these incidents were perpetrated by offenders who did not threaten, attack, or injure, we are left with 40% of DGUs in NSDS for which there is no testimony concerning specific criminal acts by the offenders. This is not to say that defenders were mistaken in their "beliefs" that a crime was intended or being committed—only that the questionnaire was faulty in not eliciting the information required to establish that a criminal victimization occurred. Kleck does not accept that "victim believed that the offender possessed a gun" in NCVS gun crimes is adequate evidence that "the gun was actually used in a threat or attack." He is not, therefore, entitled to infer criminal victimization of defenders in NSDS solely on the basis that respondents, asked a leading question ("What sort of crime did you believe was being committed against you?"), will accept one or more of seven preceded type-of crime categories. NCVS does not ask about "belief"; it asks for reports on what happened.
To sum up, the comparability of NSDS and NCVS reports of victimization is open to question in regard to the 13% of NSDS crimes not included in NCVS (including 9.5% not otherwise described "other" crimes), the 33.8% burglaries, for which there is no specific testimony about breaking and entering, and the 40% for which no specific criminal act is described. These categories are not mutually exclusive. But it would be "generous" (to use Kleck's term) to conclude that for no more than three-fifths of NSDS DGUs is comparability with NCVS with respect to victimization sufficiently secure to warrant close comparisons. NSDS by itself provides evidence (section 3.2) that could be used to raise the NCVS gun crime count from one million to 1.4 million. Hence the 1.3-1.5 million (60% of 2.14 or 2.53 million) arguably comparable DGUs are just enough to justify Kleck's "tentative conclusion" that "defensive gun uses are at least as common as criminal gun uses" (Kates and Kleck 1997, p. 175) or his earlier statement to the same effect (Kleck 1991, p. 107).
The challenge is still more daunting for anyone who would look to gun use statistics for help in inquiring, as Kleck (1997, p. xvi) recommends, "how much cost or benefit there is, and for how many people" in regard to criminal and defensive use of guns. For at least the cardinal issue, that of lethal violence (Zimring and Hawkins 1997), the substantive question is clear: "How many lives are lost from criminal use of guns or saved by defensive use of guns?" (Kleck 1997, p. xvi). His answer to the second question is unconvincing. NSDS found that 15.7% of defenders considered it "almost certain" that someone would have died had they not used guns for protection. The implied claim that 400,000 or so lives were saved by DGU "cannot be dismissed as trivial. If even one-tenth of these people are accurate in their stated perceptions, the number of lives saved by victim use of guns would still exceed the total number of lives taken with guns" (Kleck and Gertz 1995, pp. 180-81). But why one-tenth? Why not one one-thousandth? Answer: "How many of these truly were life-saving gun uses is impossible to know" (Kates and Kleck 1997, p. 230). The later discussion advisedly omits the nonsensical "If even one-tenth ...". An earlier statement (Kleck 1991. p. 129) had already taken care of the matter: "[It] will never be possible to form a meaningful ratio of genuinely comparable quantities."
As for the lesser costs and benefits, not only is it often impossible to ascertain in a survey who "offended" and who "defended," but the circumstances and nature of the gun-use events are also often obscure. In some measure this is due to failure to analyze in any depth the voluminous data already collected. Kleck and Gertz present a nontrivial estimate of lives saved in DGUs without even investigating the statistical association between perceived likelihood of death and the classification of incidents by kind of weapon (if any) the offenders carried, or by the kind of violence they directed at defenders. NSPOF collected useful information about firearms training, but the investigators did not inquire whether it is the trained or the untrained gun owners who are more likely to use their weapons in self defense. NCVS only occasionally puts out a tantalizingly brief leaflet with interesting but fragmentary data that, unfortunately, cannot easily be matched with the meager statistics on gun crimes offered in the annual victimization reports. GSS, of course, restricts its task to data collection and dissemination. The yet-to-be-analyzed information it provides about the population of OGU victims is vastly more detailed and .sociologically nuanced than any other source offers. Some of the other OGU surveys that are accessible in data archives could provide significant tidbits of data not otherwise available.
I am not holding out the hope that some definitive cost-benefit calculations will be coaxed out of archival data, seriously analyzed. But there is much to be learned from strictly descriptive studies of the character of the "gun culture" in which, according to so many of the contributors to the Dizard et al. anthology, we so uncomfortably live. Zimring and Hawkins, while arguing that in the United States the most important problem in the area of crime and criminal justice is that of lethal violence, nevertheless emphasize that counts of deaths and injuries do not suffice to measure their social costs, which include the anxiety and fear produced by the uniquely high rate of homicide and the ripple effects of these emotions on social life. Surveys are pretty good at detecting such effects (e.g., Field Institute 1994), but how anxiety and fear relate to gun use, defensive or offensive, has yet to be adequately investigated.
One positive social benefit that could be realized by a thorough evaluation of the studies in hand is the debunking of the extravagant claims now circulating about the benefits of DGUs. Dipping into just one source, the Dizard et al. (1999) collection of previously published statements, I find at least four examples of serious misunderstanding of DGU statistics. Kopel (p. 453), referring to the early report on the Hart survey by Kleck (1988), claims that "large numbers of Americans say they or someone in their households have actually used (i.e., fired or brandished) a gun for self-defense." But Kleck and, presumably, Hart provided no information on either firing or brandishing. Schulman (p. 404) states, "According to figures compiled [presumably by Kleck (1991)] from around eight different studies private citizens ... use a firearm about a million times each year to stop or prevent a crime." But none of these surveys inquired whether a crime was either stopped or prevented. LaPierre (pp. 173-74) asserts, "As many as [2.54] million crimes are thwarted each year....a dozen previous [to NSDS] studies ... show that the use of guns for defensive purposes is ... effective... because criminals flee when confronted with a firearm." None of the DGU surveys inquired about thwarting or fleeing. Kleck and Gertz (1995), who summarized the material LaPierre is referring to, cited NCVS data on effectiveness but explicitly excluded an inquiry into that question from NSDS. None of the other "previous" studies got into it either. LaPierre speculates that up to 2.5 million more crimes could appear in national statistics each year if citizens were denied the use of guns in self defense. According to NSDS data, 64.2% of DGUs are already reported to police (Kleck and Gertz 1995, Table 3; but see the authors' caution about this statistic, p. 177). Lott (p. 322) mentions "polls ... undertaken by organizations like the Los Angeles Times and Gallup showing that Americans defend themselves with guns between 764,000 and 3.6 million times each year, with the vast majority of cases simply involving people brandishing a gun to prevent attack." This is obviously based on the summary by Kleck and Gertz (1995, Table 1), where we find the 764,000 figure given for the Tarrance survey, the 3.6 million for the Los Angeles Times survey; both figures are Kleck's "adjusted" estimates, not statistics appearing in his sources. Neither of the two Gallup polls in this summary and neither the Times nor the Tarrance survey provides information about "preventing attacks" or frequencies of "simply ... brandishing." To these examples of serious misunderstandings of survey data can be added several others cited by Duncan (2000).
For a counterbalance to my one-sided catalog of errors, see Kleck's (1997, pp. 31-62) longer review of "illegitimate practices in summarizing research on guns and violence," perpetrated preponderantly, it would seem, by gun-control advocates. As Kleck observes (p. 38), "The simplest way to give a biased impression of a body of evidence is to leave out any findings the reviewer finds uncongenial."
Future surveys of gun use will have little credibility if they fail in either of two ways: by looking at only the defensive or the offensive uses or by going no further than past research in inquiring about what actually happens in these encounters. It is already common practice to have follow-up interviews by supervisors to verify the records made by the original interviewers. Why not employ as "supervisors" persons practiced and skilled in "getting the story" as the respondent understands it, as well as the set of check marks beside prefabricated answer categories? There will remain the ineradicable uncertainties as to exactly who did what and to whom. But at least we might find out whether the vitally interested parties—victims or defenders—agree with the characterizations of their experiences now being tendered by their advocates.
Table II. Regression Estimates of Percent Ever Threatened or Shot at
with a Gun as an Adult, 1994, and Annual Change in that Percentage,
Initial Year to 1994, for Men and Women, by Birth Year, General Social
Table III. Handgun OGU and DGU Rates (Percent), for California Men and
Women 18 Years Old and Over in Residential Dwelling Units, for Three Time
Periods When Incidents Occurred, Field Institute Survey, 1975
|Past 2 years||5.0||7.1||3.0||3.0||4.0||2.0|
Source: Field Institute (1976). Data were collected in personal in-home interview, November 1-6, 1975. These data were, in part, originally collected for the Bureau of Criminal Statistics, Department of Justice, State of California. The data are provided by the Field Institute. None of the above organizations has any responsibility for the analysis or interpretation of the data which the author of this article has made.
aQs 25-26: Have you, yourself, ever been threatened by someone with a handgun, again excluding military service? [If Yes] When was the last time that you were threatened by someone with a handgun?
bQs 15-16: Have you, yourself ever used a handgun, even if it wasn't fired, for self-protection or for protection of property at home, work or elsewhere, excluding military service? [If Yes] When was the last time that you used a handgun for protection?
c(l) Excludes and (2) includes "Don't remember" time of last incident.
Table IV. Actions by Offenders in Gun Crime Incidents, NCVS 1993
Survey, with Comparative Summary Data for Two Earlier Periods
|Action (Response categories, Qs 28 & 29)||Recode||(Unweighted)||of Total|
|Shot (29: 4)||(1.1)||14||2.2|
|Shot at (28: 8; 29: 5)||(1.2)||52||8.3|
|Verbal threat to kill ((28: 2)||(2.1)||146||23.3|
|Rape/sexual assault, including threat, attempt,- unwanted sexual contact (28: 1,4,5,6; 29:1,2,3)||(2.2)||14||2.2|
|Attack/attempt with weapon, object, except gun (28:9,10,11:29:6,7,8,9,10,11)||(2.3)||32||5.1|
|Manhandling (28: 12,13:29:12,13)||(2.4)||66||10.5|
|Other verbal threat (28: 3)||(2.5)||55||8.8|
|Other action (28:14; 29:14)||(2.6)||12||1.9|
|Weapon present or threatened with weapon (28: 7)||(3.0)||236||37.6|
|Summary (Percent of total)||1979-87||1987-92||1993|
Source: Handgun victimizations 1979-87 and 1987-92, U.S. Bureau of Justice Statistics (1990, 1994a). All gun incidents 1993, unpublished tabulation from public use files of the U.S. Department of Justice, Bureau of Justice Statistics, National Crime Victimization Survey, 1993. Summary categories but not detailed action recodes are comparable across time periods. Recode is hierarchical: each category may occur with any of the lower ones. In particular, all categories are accompanied by "gun present."
Figure 1. Percent Threatened or Shot at with a Gun as an Adult, for Men and Women in Cohort Born 1950-54, by Year of General Social Survey
Figure 2. Percent Threatened or Shot at as a Child, for Men and Women, by Year of Birth, General Social Survey