This site is copyrighted, supported, and maintained by the Equal Justice Foundation.
| EJF Home | Join the EJF | Comments? | Get EJF newsletter | Newsletters |
| DV Home | Abstract | Contents | Tables | Index | Bibliography |
| Chapter 4 — Domestic Violence Statistics |
| Next — Reading between the numbers |
| Back — Processes explaining the concealment and distortion of evidence on gender symmetry in intimate partner violence by Murray A. Straus, Ph.D. |
Use of statistics in sociological studies
Factors to be considered when looking at statistics
Statistics through a feminist lens
Telephone surveys and statistics
Tabulation of statistical factors
Abstracted from the Irish Times, September 18, 2000
One of the things that is extremely frustrating in dealing with sociological issues such as domestic violence are the presentation of the statistics.
Imagine a survey about relationships between cats and dogs that surveyed only cats. Inevitably, it would suggest that all conflict between cats and dogs is caused by dogs.
The conclusions would state unequivocally that cats are always “victims.”
Therefore, to end the “battering” dogs must be banned from the house.
Dog lovers would protest that cats are deeply duplicitous creatures who feign innocence but are actually the cause of most of the difficulty. They would say that though dogs are demonstrably stronger than cats, most dogs are extremely gentle in their treatment of cats.
If anything, in the opinion of dog lovers, it is cats that seek to take advantage of their alleged weakness to torment and provoke dogs.
Uncommitted observers, having no bias towards either cats or dogs, would immediately point out the fallacy of a system of statistical analysis that excluded data from one of the parties to the alleged discord.
Invoking such an analogy with domestic violence, however, is likely to provoke pseudohysterical complaints about “blaming the victim” from those with interests in this issue. But the above analogy succinctly objectifies the current thinking in the domestic violence industry.
Denzin (1980, p. 213) has stated that in order to prove a causal relationship in sociological studies three properties must be demonstrated:
1. Association: Proof of significant associations or correlations of the key variable must be shown. The researcher must show the causal variables in the “symptom” produced variations in the dependent variable.
2. Time: A clear temporal relationship, wherein one factor precedes the other, must be shown. The researcher must show that the causal variable occurs before the dependent variable.
3. Intervening variables: An explanation of the relationship of intervening factors as catalysts or products must show that this causal system is not spurious — not the product of other variables.
Rather than meeting these conditions, one gets the impression instead that most authors dealing with family violence have had, at best, a single course in “business” statistics. The mean is possibly understood but standard deviation is a bit beyond them. Such things as determining a representative population or multivariate analyses are techniques they have never heard of, let alone attempted to gather the data required, though they often seem quite content to stuff such data as they do have into a computer program such as SAS or SPSS. The output of the program is then regarded as gospel, blithely ignoring the GIGO principle — Garbage In, Garbage Out.
I have also become aware of another vital difference in sociological studies as compared with the physical sciences. In sociology it is common to start out with a fixed hypothesis and then seek data to support that idea. Frequently the original hypothesis becomes an idée fixe. That obsession is particularly noticeable in feminist literature. Conversely, in the physical sciences one is confronted with a bewildering array of data for which one has multiple working hypotheses. The problem then becomes (1) a process of elimination of unworkable solutions and (2) collecting ever more data that commonly refutes the original hypotheses altogether. Sociology would do well to adopt the methodology of the physical sciences as human problems almost never have a simple solution.
There are lies, damn lies, and statistics.
In reading the work of those who see domestic violence through a feminist lens the following errors constantly recur:
• Their assumptions never seem to be defined.
• Correlation is consistently confused with causation.
• Ensuring the data are a random sample from a representative population is considered a trifling detail they can ignore.
• Control or comparison groups are unheard of in clinical studies.
• Error estimates are never presented.
• Numbers are posted with a precision totally unjustified by the accuracy of the underlying data.
• Propaganda, or agitprop, substitutes for a hypothesis.
• Negative data are ignored, or condemned on ideological grounds, i.e., men are all violent beasts who batter women to uphold the patriarchy.
Yet on the basis of such poor statistical analyses, and a few subjective experiences often presented as anecdotes, laws have been passed denying the constitutional protections evolved from centuries of hard-won experience and defined by geniuses far greater than we.
One of the best exposés of the misuse of statistics is the August 24, 1996, editorial in the Canadian Globe and Mail by Margaret Wente on the $4-billion abuse fiction that reviews economic estimates of the cost of violence against women. For example, after obtaining a copy of the original report, she states:
“The exact estimate given in the report is $4,225,954,322. This number is the product of multiple extrapolations, a persistent confusion of correlation with causation, monetization of the non-monetary, wild guessing, weird logic and counting everything imaginable.”
Yet it is estimates of economic costs such as these that Congress used to justify invoking the Commerce clause of the U.S. Constitution in order to pass the Violence Against Women Act. Fortunately, the U.S. Supreme Court may be taking a dim view of Congress' fascination with using the Commerce clause as a basis for an ever expanding series of federal crimes and in June, 2000, found portions of VAWA unconstitutional. However, that did not deter the Congress from passing VAWA II a few months later.
One of the major family violence studies was the National Violence Against Women Survey (NVAW) survey. Much has been made of the data from this survey that was conducted entirely by telephone with a sample size of 8,000 women and 8,000 men (who were originally to be excluded). The broad problems with the NVAW survey are addressed in Straus (1999). Here we deal with the problems and biases of telephone surveys in general.
First, of necessity the sample population for telephone surveys is drawn from published telephone books. Therefore, everyone without a landline telephone, with an unlisted or unpublished phone number, use only a cell phone, or who appears on such no call lists as Colorado maintains, is immediately excluded from the survey. Consider that citizens who have been involved in domestic violence are well advised not to publish their phone numbers and most don't.
Second, millions of people presently have caller ID, particularly if they have been involved in family violence or have been stalked, and will not answer calls from numbers they don't recognize. Call blocking, particularly for anonymous callers is also widely used. Even if the call makes it through, many people use their answering machines for call screening. As pollsters typically do not identify themselves or leave messages on answering machines (who would call back and say they wanted to answer questions for a phone survey?) an estimated one-quarter to one-third of those with published phone numbers are excluded by these factors.
Third, only about 20-50% of people who the phone survey does reach will participate. An overview of the problems with response rates for telephone surveys has been published by the American Association for Public Opinion Research for those who care to delve more deeply into this problem.
Of those few who do answer their listed telephones, with no call screening or blocking, how many are willing to honestly bare their souls to a total stranger on a subject with the emotional impact of family violence? Thus, even those few who do participate may answer “Don't know” or refuse to answer some survey questions, or they will simply lie. For example, a woman who blatantly used a restraining order entirely to gain advantage in a divorce will almost certainly claim in a telephone interview that she was abused.
These factors eliminate all but about 10-20% of the general population. Telephone surveys are thus hardly random and representative. Considering these factors it is hardly surprising that, as Straus (1999) notes, the NVAW survey found only: “...1/12th of the rate obtained by family conflict studies such as the National Family Violence Surveys.”
A fundamental theorem of statistics
In looking at statistics about domestic violence and abuse there are many factors that must be sorted out to ensure the data are relevant to the problem as stated formally above. Most of these are simply common sense and the average individual understands the principles intuitively once the jargon is cleared away:
1. Correlation does not imply causation. A fundamental theorem of statistics but often ignored.
For instance, the relative numbers of arrests of men and women for domestic violence are used as a basis to claim that men are most commonly the batterers. However, such numbers have nothing whatsoever to do with who initiated the violence or was the abuser. All these numbers say is that the police more commonly arrest men when called with a report of a domestic disturbance. The reasons for that arrest policy are manifold but have no bearing on who was the violent party in the relationship.
Or take another example, an interviewer asks if the woman has ever been hit or bruised by their partner? A positive answer does not imply domestic violence if the couple practices BDSM. Or they may simply play very roughly with one another and she got bruised accidentally.
The question: “Has your partner ever forced you to have sex?” does not imply rape if the couple are into bondage and domination. The recent National Violence Against Women Survey (NVAW) survey appears to contain such errors.
2. Do the studies include random data from the entire population, both men and women? Virtually all feminist studies include only data from or about females. Thus, they ignore at least half the problem as did the Canadian study reviewed by Maguire. Oft times subjects in feminist studies are not randomly chosen, or are self-selected, as in a mail-in survey from Cosmopolitan magazine readers.
Repeatedly, examples are found where surveys either do not ask men questions about assaults by their female partners, or the data are collected and then blatantly ignored. In other instances the data are compromised by reporting or system biases. For example, the National Violence Against Women Survey did not ask women about their assaultive behavior against their intimate partners.
Even Lenore Walker (1979, p. xiii), who has written extensively about battered women, shares the limitations of her own studies when she writes, “These women were not randomly selected, and they cannot be considered a legitimate data base from which to make specific generalizations.”
As noted previously , John Maguire has detailed how Canadian national policy was based on a survey that simply ignored the data on males that was collected. In the latest Alberta, Ontario study, only the statistics which pertained to female victims of domestic violence were presented to the Ontario government. The data presented indicated that 12.9% of the men in the study behaved violently toward their spouse. In 1999 the data were reexamined and it was noted that the study also showed that 12.5% of the women behaved violently toward their spouses. The study also suggested that women were almost twice as likely to “hit or try to hit” their spouses, 9.0% of the wives compared to 5.4% of the husbands. The government officials never saw these statistics, but nonetheless authorized $858,000 for an “advertising campaign featuring the slogan: Wife assault is a crime. There's no excuse.” (Laframboise, 1999)
In a study by the Kentucky Commission on Women, Straus (1997, p. 212) writes that researchers “intentionally suppressed” information that “38% of attacks were by women on men who, as reported by women themselves, had not attacked them.”
Such actions by investigators are criminal and represent a complete lack of honesty and integrity. Yet many, if not most, of the current laws regarding domestic violence and abuse are based on such fraudulent, if not criminal practices.
3. Context. Do the studies only include factors that are a valid measure of the subject? For example, feminists are very fond of using crime statistics because they show women assaulted at a much higher rate than men. However, under current laws the reality is that even if a man does call the police to report he has been assaulted by his mate, there is a very high probability that he will be the one arrested and charged. First, the crime statistics then count this as an assault on a woman. Secondly, many men are aware of the gender bias in law enforcement. Knowing that if you call about a female assault, the police are more likely to arrest you, why would a man call them?
There is also evidence that women are seven to ten times more likely than a man to report an assault for the same level of injury. Even with homicides it is less likely that a woman killing her intimate partner will be counted as domestic violence, e.g., it may be counted as self defense or she hired someone to do the killing as Warren Farrell points out in his Twelve Female Only Defenses. As a result crime statistics on family violence are useless. However, the recent National Violence Against Women Survey (NVAW) makes exactly these errors.
Tjaden and Thoennes (2000, p. 23) explicitly state: “...it is likely that the manner in which screening questions are introduced and framed has more of an effect on intimate partner victimization rates than does the overall context in which the survey is administered.” when attempting to explain the extensive differences between their survey and the work of Straus and Gelles (National Family Violence Surveys) and the National Crime Victimization Survey (NCVS). Of course, Tjaden and Thoennes argue that we should believe them but not everyone else.
4. Conclusions are based on circular arguments. For example, police training is currently based on statistics showing that most DV arrests are of males. Therefore, police are instructed that when evaluating who is the primary aggressor in a domestic situation it is likely the male. So the man should be the one arrested, reinforcing the statistics that most arrests are of males.
5. Focus. Do the studies concentrate on a particular topic or wander all over the feminist political agenda? If you are studying family violence then stalking cannot be part of the study because, by definition, stalking does not occur when a couple are living together and often does not involve intimate partners.
Unfortunately, the focus of many feminist studies isn't on obtaining factual data about the problem of domestic violence and abuse. As previously discussed, advocacy research is a common tool used to substantiate their underlying, and unstated, objective of demonizing men. They want to fix the blame, not the problem.
6. Are the studies rate dependent? That is do they have time as a variable such as “...per year” or do they cover the entire period a couple are together? If a couple are together 20 years, and they have one shouting and shoving match during that time, that is a very low rate.
However, if time isn't a function of the survey, and contained in the questions asked, then the couple should only be counted as having experienced family violence sometime in their relationship.
Note that the term rate is commonly used where the investigator seems to mean ratio. From the sample questions, the 1995 National Crime Victimization Survey (NCVS) appears to contain this error.
7. Size of the survey. For the year 2000, the Census Bureau expects the population of the United States to be about 202 million men and women over age 18. Thus, a sample of a few hundred couples will have large errors with regard to the behavior of the entire population. Typically, the error bars are unreported though they are easily calculated from statistical methods and programs like SAS and SPSS generate this information. Most inconsistencies between surveys are caused by sampling errors and if error bars were shown it is likely the differences are within the margin of error for each survey.
Straus (1993, p. 77) also warns about what he terms the “representative sample fallacy” where the community sampled contains very few cases of the variable the survey is attempting to measure. The NVAW survey of Tjaden and Thoennes (2000) is an excellent example of this fallacy where they are reduced to attempting to extrapolate to the general population from a sample size of as few of five cases.
8. Are participants in the sample randomly chosen? If you only ask females in a shelter for battered women, or who are in a divorce and child custody suit, you will not get answers that represent the general population, or males at all. Yet many feminist surveys basically do this and current laws are based on such biased results. Straus (1993, p. 77) refers to this problem as the clinical fallacy.
Feminists make much of the high percentage of injuries reported by women as compared to men. However, studies consistently show that for the same level of assault, women are 7 to 10 times more likely to report an injury or call the police. That constitutes an overwhelming reporting bias when police or hospital figures are used for statistical analysis. In short, men don't tell.
Note, however, when the reverse is true, and women don't tell in rape cases, radical feminists are equally upset. There they want to ignore the police and court statistics and rely solely on the social surveys they so blithely ignore for domestic violence.
9. What percentage of the people asked answered the survey? If you, or someone you know, has been involved in domestic violence you are much more likely to participate but your answers will skew the results.
Ask yourself how many men do you know who would participate if they answered the phone and were asked to spend a half hour or more answering questions for a “National Violence Against Women” survey? Click, as the phone is hung up, is the most likely response, even from many women.
Presently only about 20-50% of people asked will participate in any given survey. As a result, surveys commonly do not include many people without a direct interest in the subject. Further, even those few who do participate may answer “Don't know” or refuse to answer some survey questions. Such answers are typically excluded from sample analyses, most commonly due to poor survey design.
People who simply refuse to participate are almost never counted in the results. Their non-participation further skews the results. When 80% or more of the people asked refused to answer the questions the survey is not random. It is also very unlikely that those who don't answer the questions would agree entirely with those who do. The results are further skewed if you title the survey something like “National Violence Against Women” that few sane men would consider participating in.
10. What questions were asked and how and when were they asked? If you ask females emerging from a battered woman's shelter: “Do all men deserve to be shot out of hand?” you are likely to get a positive response from a significant number. Would a significant number of randomly sampled women answer “Yes” to that question? Almost certainly not. However, if you ask a random sample of women whether men who beat their wives should be severely punished, the likely answer is: “Yes.” Ask the same women how females who assault their partners should be punished and the probable answer is: “That doesn't happen!” These three examples show how such studies commonly produce nonsense results.
One common method of biasing domestic violence surveys is to ask men and women different questions, or not ask men some of the questions women are asked or vice versa. Example: A man might be asked how often he hits his wife. But the wife won't be asked how often she hits her husband though the wording is likely to be much more subtle in the survey.
Another factor is whether the man and woman were asked the questions independently of one another or while they were together. The NCVS interviewed the couples together. That probably makes a man or woman reluctant to respond honestly and openly about his/her abuse. The NCVS was also presented to respondents as a “Crime Survey.” People may only report domestic violence as a crime if it is very severe or chronic. Pushing or shoving during a family argument is unlikely to be considered a crime and would go unreported.
As noted under context above, Tjaden and Thoennes (2000, p. 23) state this point explicitly in trying to explain the substantial differences between their NVAW study, the NCVS survey, and the many NFVS surveys by Straus, Gelles, Steinmetz and others.
11. What is the study attempting to measure? Or, define the problem before attempting to solve it. For example, if we measure violence by actions then women appear to be more violent than men in a domestic environment. Other measures, such as crime statistics, will provide different results.
Because of physical differences females are thought to be more likely to suffer injury regardless of who initiates the violence. Women are also 7 to 10 times more likely to report their victimization by domestic violence to police than are men and children. These physical and behavioral differences inflate figures suggesting women are exclusively “victims.”
One also commonly encounters the use of terms like “abuse” or “injury” without any clear definition of what constitutes such. A man is very unlikely to seek medical help for a bruise or a scratch but women frequently will. So is a bruise or a scratch an “injury” and treated equally for both sexes? If not, the survey data are badly biased by the lack of definition.
There is also the problem of the point of measurement. If a woman repeatedly assaults her male partner until finally one day he decks her, in feminist studies that is typically measured solely as violence against the female. In more balanced studies that would be counted as female on male violence or mutual combat.
There is nothing inherently wrong with such biases but they should be clearly defined and measured. However, advocacy research takes such biases to an extreme and tends to discredit the study for all but true believers.
We take the position that actions are the proper measure and view the arguments from a male perspective.
12. Mixing the data: When one asks the question, “What is the percentage difference between female and male victims of domestic violence?” most feminists will go to archival data, such as the National Crime Victimization Survey (NCVS), and say that about 95% of victims are women and 5% are men. Then if one asks, “What is the projected number of female victims of domestic violence in the U.S.?” groups such as the National Coalition Against Domestic Violence (NCADV) will usually go to survey data other than the NCVS, such as studies based on the Conflict Tactics Scale, to claim that two million women a year are seriously assaulted in the United States every year.
Clearly, data are selectively pulled from whatever source favors the feminist viewpoint on the topic of domestic violence and abuse. That is particularly true with rape data. There feminists want to ignore the law enforcement data in favor of advocacy studies that show much higher rates and broadly define the meaning of rape.
In effect, they are comparing apples with oranges. Here we attempt to present a consistent picture from all the available data with the intent of understanding and fixing the problem of family violence.
13. What other factors may be influencing the study results? For example, to date we have not found any attempt to subtract from any domestic violence studies the estimated 10-15% of couples who practice some level of BDSM in their relationship.
Another example: Prostitutes may be using battered women's shelters to escape from their pimps, yet they are counted in the shelter statistics. Also, prostitutes who assault their customers are not counted in intimate partner violence surveys though such assaults meet all current definitions of DV assault.
Human relationships are complex and often not subject to rigorous mathematical analysis. Many behaviors now branded “domestic violence” by feminists are well within the range of normal human behavior. For example, couples often scratch and bite during and before sexual congress. Alex Comfort, in his widely read book The Joy of Sex, pointed out in 1972 that: “Tenderness does not exclude extremely violent games...”
In his Table 3, Straus (1999) shows how and why statistics are selectively used by service providers and feminists as compared with academics and researchers.
Extensive analyses of statistical data are provided by David Fontes in his Violent Touch article.
It is natural for differing sides to use statistics that favor their cause. Unfortunately, only rarely have feminists been statistically rigorous. For example, there is no evidence at all of such rigor in the book The Battered Woman by Lenore Walker (see review by Sheaffer ), probably one of the most influential feminist works in Colorado and the United States.
| EJF Home | Join the EJF | Comments? | Get EJF newsletter | Newsletters |
| DV Home | Abstract | Contents | Tables | Index | Bibliography |
| Chapter 4 — Domestic Violence Statistics |
| Next — Reading between the numbers |
| Back — Processes explaining the concealment and distortion of evidence on gender symmetry in intimate partner violence by Murray A. Straus, Ph.D. |
This site is supported and maintained by the Equal Justice Foundation.