The Sociological Study of Suicide: Methodological Issues

Two broad sociological orientations guide the methodological inquiry in the discipline. The first the current paper labels “microsociology”; the second is called “macrosociology.” These two broad orientations can be distinguished along a number of dimensions, including the unit of analysis, the research method employed, and their chief sources of knowledge or data sources.

Microsociological research on suicide uses the individual as its unit of analysis. The key research method is the interview or “psychological autopsy.” Sometimes existing data sources are employed for an analysis of a set of case histories, such as the half-dozen cases pulled together by Douglas (1967, pp. 284–340) in his call for microsociological work on suicide. Jacobs's (1971) study of approximately 40 adolescent suicide attempters follows this model. A more recent example that employs survey research techniques on a much larger sample of individuals is the work by Maris (1981). The statistical techniques in this stream of research range from none at all (Douglas, 1967) to simple bivariate tables (Jacobs, 1971) to sophisticated quantitative techniques such as the multiple-classification analysis found in Maris's (1981) study. Hence, the mode of statistical analysis does not distinguish this approach to the study of suicide; rather, its unit of analysis, the individual suicide victim or attempter, is its distinguishing characteristic.

The second mode of analysis is represented by research on large population aggregates. The focus is not on the meanings associated with the suicide of the individual actor, but on the characteristics of population groups, as a whole that might increase the probability of suicide in the group. Units of analysis here include cities, standard metropolitan statistical areas (SMSAs), nations, the 50 United States, the 3,000 U.S. counties, and other population aggregates. Such research can assume a cross-sectional or a longitudinal strategy wherein one studies suicide statistics at one point in time (e.g., Bankston, Allen, & Cunningham, 1983; Barnes, 1975), or over a time frame for a population. The latter would include studies of the United States as a whole (e.g., Pierce, 1967, for 1919–1939) and research on other nations (e.g., Stack & Danigelis, 1985, for 1919–1972 for each of 17 nations).

In recent years there has been some variation in a third unit of analysis, time. While classic works on suicide have almost always used the year as their time frame, in the last 10 years there has been a series of macrosociological studies using the month and even the day as their units of analysis for time (e.g., Baron & Reiss, 1985; Phillips, 1974). These departures from tradition have been largely due to a need to test the imitation theory of suicide. In order to pinpoint the triggering effect of suicide stories on additional suicides, it is important to have a shorter time frame than a whole year.

The principal research method employed in macrosociological research is based on existing or archival data relating to populations, not individuals. Here the researcher does not interview individuals involved in suicidal behavior or their significant others; instead, the researcher is apt to spend his or her time in the library digging out data on such variables as rates of suicide and unemployment. The data sources are often U.S. Bureau of the Census volumes and other government publications providing a wide variety of data tables. The variation in the rate of suicide for an aggregated population is explained by indicators of suicidogenic conditions such as unemployment, divorce, and low religiosity (e.g., Danigelis & Pope, 1979; Pope & Danigelis, 1981; Stack & Has, 1984; Trovato, 1986).

The overwhelming tendency in macro approaches to suicide is to rely on official suicide statistics drawn from various sources in local and state governments and the federal government. Chief data sources include the U.S. Public Health Service's (1964–1985) annual publication, Vital Statistics of the United States, and, more recently, data tapes supplied by the same source that contain records on all deaths in the United States for the period from 1968 to 1983 (thus far). From these tapes one can extract over 375,000 computerized death certificates on suicide victims. Sociological researchers have been able to produce data sets based on these magnetic tapes, such as daily counts of suicide, and age-, race-, and sex-specific monthly counts (e.g., Baron & Reiss, 1985; Bollen & Phillips, 1982; Stack, 1986c). In the case of age- and gender-specific daily and monthly counts, these tapes offer the only avenue to constructing such data sets.

Some of the recurrent independent variables used in the macro-sociological tradition include economic conditions, such as unemployment and industrialization (Henry & Short, 1954; Stack & Danigelis, 1985; Stack & Has, 1984); political factors, such as war and presidential elections (Marshall, 1981; Wasserman, 1983); and a wide variety of social conditions of group life. The last-mentioned include divorce (Stack, 1980b; Wasserman, 1984b); religiosity and religious integration (Pope & Danigelis, 1981; Stack, 1983a, 1983b); suicide stories in the mass media, which are thought to trigger additional suicides (Phillips, 1974; Phillips & Bollen, 1985; Stack, 1986c); the process of modernization and its subcomponents, such as urbanization (Quinney, 1965; Stack & Danigelis, 1985); the uprootedness associated with migration rates (Stack, 1980a); and status integration (Gibbs & Martin, 1964; Stafford & Gibbs, 1985).

The purpose of the present paper is to address three current methodological issues within the context of the macrosociological approach to the study of suicide. The choice of the macrosociological approach over the microsociological approach is somewhat arbitrary. However, most of the work in the area assumes a macrosociological orientation (see Stack, 1982, for a review). This gives the current paper more studies to draw on. The three issues are not an exhaustive list of current problems in the literature; instead, they are meant only as a more or less representative set of issues facing people in the field in the 1980s. The issues are (1) the measurement of suicide rates and independent variables related to suicide; (2) the methodology related to testing the imitation/suggestion theory of suicide; and (3) the need for comparative research. Wherever possible, the current paper attempts to develop some methodological strategies for attacking the existing problems that often retard progress in the field.

The Measurement of Suicide and Independent Variables Related to Suicide

First, there are numerous issues related to the measurement of the basic phenomenon under study, suicide. In addition, this first section explores measurement issues regarding three core sociological variables: economic stress/anomie, religion, and political conditions.

Studies in the macrosociological tradition have continued to rely, for better or worse, almost exclusively on official suicide statistics as their data base (e.g., Baron & Reiss, 1985; Phillips & Bollen, 1985; Stack, 1985; Stafford & Gibbs, 1985). Classic criticisms of these official statistics (e.g., Douglas, 1967) have often argued that they are quite tenuous, given the alleged widespread concealing of suicides; such alleged concealment may lead to a misrepresentation (specifically, an undercounting) of suicides.

The first really systematic test of the validity of cross-sectional official suicide data was performed by Pescosolido and Mendelsohn (1986). The authors collected data on suicide for 404 U.S. counties with populations over 250,000. They were able to distinguish between two sets of independent variables: social construction and social causation factors. Social construction variables are ones thought to be associated with systematic misreporting, which is misreporting whose rate varies according to the nature of the medical-legal system responsible for classifying deaths. The medical-legal system includes several variables measured in the Pescosolido and Mendelsohn study: percentage of a county served by a medical examiner system, percentage served by an elected coroner or medical examiner, percentage served by a coroner who can call an inquest, and other indicators of the resources available for classifying the cause of death. The latter include the percentage of the population of a county with a local toxicologist and/or pathologist (Pescosolido & Mendelsohn, 1986, pp. 85, 98). The results indicated that while systematic misreporting was found to exist across different medical-legal systems, such misreporting had little discernible impact on the relationships between social causation variables and suicide. That is, after controls for the variation in medical-legal systems were introduced into the equations, there was little change in the effects of education, relative income, divorce, migration, religious composition, and other independent variables on the rates of county suicide.

A limitation of the Pescosolido and Mendelsohn (1986) defense of official data is that it is based only on cross-sectional U.S. data. It is not clear whether or not their results would be replicable over time or longitudinal data for the United States. In addition, replication studies for other nations would be useful for the purposes of defending the results of cross-national studies based on official data.

The problem of taking disguised suicide into account in empirical analyses has also been the subject of a series of papers wherein single-car motor vehicle fatalities have been used as an index of parasuicide (Bollen & Phillips, 1981; Kessler & Stipp, 1984; Phillips, 1979, 1982). Given that these deaths have a suicidal component, one would assume that their rates would respond to the same etiological conditions underlying suicide. This has generally proven to be the case. However, the work to date for the United States, unlike that for England, has restricted its aggregate measures of parasuicide to such “autocides.” The current paper proposes that researchers begin to follow the example of British suicidologist Brian Barraclough (Barraclough & White, 1978) and incorporate other measures of parasuicide into their models.

Using the current international classification scheme for deaths in 1979, Table 1 gives the numbers of deaths in the United States for parasuicidal categories. Work is needed on such forms of parasuicide as violent deaths from unspecified causes, accidental falls, and accidental drug-related poisonings. Other possible unexplored measures of deaths with a suicidal component would include accidental drownings (n = 5,678 in 1979) and deaths caused by accidents with firearms (n = 2,004 in 1979).

Actually, given that the total number of official U.S. suicides was 27,206 in 1979, the suicide rate would be only approximately double its current rate even if all of the standard forms of parasuicide were covered-up suicides (and that is quite unlikely). The official data, then, may be more accurate than the critics claim.¹

The discussion now turns to a series of issues involving three core independent variables. A time-tested condition in the etiology of suicide is the broad category of economic stress. This is often interpreted from a Durkheimian standpoint through the concept of economic anomie (Brenner, 1984; Durkheim, 1897/1966; Platt, 1984). Much of this work has focused on more or less objective measures of a gap between economic goals and economic realities, such as the rate of unemployment, average duration of unemployment, business cycle indicators, and so on (see Platt, 1984, for a review). However, these measures tend to measure only the economic means portion of the means-versus-ends process of anomie (Ahlburg & Schapiro, 1984, p. 98).

However, recent work on labor market conditions has called attention to a new form of economic stress: underemployment (Easterlin, 1980; Jones, 1981). Therein, it is not only joblessness per se that increases suicide potential, but other factors as well, including being underpaid and overeducated for the job one performs. A popular example of underemployment is the PhD who ends up driving a taxicab. This problem is said to have increased dramatically in recent years in many industrialized nations, given the glut of college graduates on the labor market. This has been especially true in those years when the postwar baby boom cohort hit the labor market (Easterlin, 1980).

While hard data on such underemployment are difficult to find, substitute measures can be developed that capture the economic stress of young people. One very promising index has already been found to be closely linked to youth suicide in the United States: relative cohort size (RCS). A common measure of RCS is the ratio of young persons (e.g., 15- to 29-year-olds) to older persons (e.g., 30- to 64-year-olds). This simply measures the size of the youthful cohort relative to the older cohort, but the ratio is also taken as a stress index for the younger cohort. Ahlburg and Schapiro (1984) found a strong association between changes in the ratio of 15- to 29-year-olds to the number of 30- to 64-year-olds and the incidence of suicide. As this ratio increases, it is contended that population rigidities appear in the labor market, resulting in a deteriorating economic position for youth; this is associated with increases in suicidal behavior. In contrast, as the ratio increases, the labor market position of the middle-aged cohort improves (given their relatively small numbers in the labor force), and their suicide rate does not increase.

Ahlburg and Schapiro (1984, pp. 98–100) contend that the RCS variable is preferable to traditional measures of economic stress, such as unemployment, relative incomes, and the like. The RCS variable captures not only the means available for economic success, as measured through its association with the income gap between the young and middle-aged; it also captures the aspirations of the young. That is, it captures the socialization experience of the young. Parents from a small cohort produce children who have relatively high income aspirations, reflecting the favorable labor market of the parents. Hence, the RCS variable correctly takes into account the full sweep of relevant social behavior, both economic means and aspirations.

The RCS effect may be especially strong in nations such as the United States, where college enrollments relative to population size are among the highest in the world. Mass higher education may interact with RCS to increase suicide potential through raising students' expectations for success. Today's youths are the best educated in our history, yet real starting salaries in most fields have been falling since the 1960s. That trends in RCS are associated with trends in youth suicide can be seen in an inspection of American data. Between 1954 and 1978, RCS increased steadily from .49 to .72, while the suicide rate for 15- to 24-year-olds also increased steadily from 4.2 to 12.6 per 100,000. Since the late 1970s RCS has declined, and there has been a parallel leveling-off or decline in youth suicide rates (Stack, 1986b).

Some caution, however, needs to be exercised in interpreting the RCS-suicide relationship. It has not been rigorously established that the RCS variable affects individual-level behavior. While it is known that underemployment affects individual-level behavior (e.g., B. H. Burris, 1983a, 1983b; V. Burris, 1983), it is not certain exactly what individual-level mechanisms account for any relationship between RCS per se and such individual-level behaviors as suicide, crime, and personal alienation. For this reason, many writers may wish to continue to use more or less traditional macroeconomic indicators of economic stress, possibly supplemented with some modifications. For example, an employment adequacy index might be constructed through adding together those who are unemployed and those who are inadequately employed, including those who are paid below a poverty-level wage.

However, one should not be overly cautious in applying the RCS variable to suicide data. There are some micro-level data indicating that persons at a disadvantage in the RCS phenomenon have a relatively high level of suicidogenic moods. Possibly the most convincing micro-level evidence to data was discussed by Veroff (1978). Veroff compared the results of two surveys of the mental stress of young adults. One was done in 1957, a period in which we would expect the young to have benefited from a relatively small generation size. The other was conducted in 1976, a year in which we would expect the young to have suffered from a relatively large generation size. The data showed a marked increase in the mental problems of the young. For example, the percentage who reported worrying “a lot” or “always” rose from 32% to 51%. The percentage reporting that they were “very happy” declined from 39% to 31%. Other work has found linkages between the related concept of underemployment and suicidogenic psychological problems, including low self-esteem, depressed mood, and low life satisfaction (Mandilovitch, 1977; see V. Burris, 1983, pp. 465–466, and Easterlin, 1980, p. 189, for reviews). V. Burris (1983) argues that the main consequence of overeducation-underemployment in the American context is the worsened mental health of those involved-not political movements for change, and not even job dissatisfaction per se.

To turn to religion and suicide, classic work used religious affiliation as an approximation of the degree of religious integration, wherein Catholics were viewed as more integrated than Protestants (Durkheim, 1897/1966). Recent work has challenged the notion that religious affiliation has much to do with suicide potential (Bankston et al., 1983; Pescosolido & Mendelsohn, 1986; Pope & Danigelis, 1981; Trovato, 1986). Other writers have, however, found moderate to strong associations between the religious factor, broadly conceived of as “religiosity,” and the suicide rate. Indicators of religion such as church attendance, church membership rates, and even religious book production have been found to be associated with lower rates of suicide (Breault & Barkey, 1982; Stack, 1983a, 1983b; Stark, Doyle, & Rushing, 1983). Given these new findings, several writers contend that it is time to move beyond Durkheim's theory and operationalization of religion and to use indicators of religiosity instead (Stack, 1983b, 1985; Stark et al., 1983).

These new measures of the religious factor may be more important predictors of suicide potential than religious affiliation for a number of reasons. First, as I have argued elsewhere (Stack, 1980c) at some length, differences between Catholics and Protestants have narrowed considerably. In contrast to those of Durkheim's day, today's Catholics are quite similar to Protestants on a number of important dimensions of religion, including ritual attendance, strength of orthodoxy, and so on. Given a convergence in the strength of religion between the two groups, a similar convergence may be anticipated in their suicide rates. Second, however, while religious affiliation may be insignificant in the late 20th century, this does not mean that religion per se is inconsequential to the incidence of suicide. As Stark and Bainbridge (1980) point out, religion assuages all manner of human disappointments—death, unemployment, poor health, divorce, and so on—for the seriously devoted population. Suffering is more readily taken in stride by those who believe that heavenly glory will be the reward for their earthly sufferings. (For additional theoretical arguments on the cushioning effect of religion on stress, see Stack, 1983a, pp. 364–366.) Therefore, the critical distinction for suicide researchers may be to differentiate the religious population from the relatively nonreligious population. Indicators such as church membership rates may be tapping the strength of religion in various populations, which in turn tells us something about the extent to which a population can cope with stress. Religion can be viewed, in part, as a set of powerful coping mechanisms that involve the supernatural.

Recent methodological issues have also been at the center of the debate over the impact of political variables, such as war and elections, on the suicide rate. Recent work in this area has introduced new methodological advances that have overturned the traditional wisdom on the variables under investigation. However, I would argue that still more work is needed before we can conclusively reject the traditional Durkheimian arguments related to political factors.

Marshall's (1981) path-breaking work on the effect of international wars on U.S. suicide rates found that the real reason why suicide goes down in the United States during wars is improved economic conditions, not social integration. Previous analyses that failed to control for the improvement in the labor market brought about by a war economy had associated the decline in suicide during popular wars with social integration. It had been contended that wars reduce suicide, since the common people feel more integrated as they bind together against a common external threat. However, controlling for the rate of unemployment, Marshall found no relationship between war and suicide. This indicates that war reduces suicide through reducing economic stress/anomie rather than through promoting social integration.

There are some unanswered questions on Marshall's (1981) findings. First, there is the possibility that war might have a lagged effect on suicide. That is, in the aftermath of a great international war involving millions of troops, such as World War II, there is often a period of social disorganization—note the rise in divorce to a record level in the late 1940s. Furthermore, defeat in a war may put a population in a suicidogenic mood.

Boor (1981) presented evidence of a presidential election “death dip,” wherein suicide goes down just before presidential elections in the United States. However, Wasserman (1983), using more rigorous econometric techniques, found that this relationship was spurious. Unemployment also tends to decrease in the months before an election; once this factor is controlled for, the relationship vanishes between suicide and elections.

The Mass Media and Suicide

The last 10 years have brought about a rebirth in the assessment of the influence of the mass media on suicide (Baron & Reiss, 1985; Bollen & Phillips, 1981; Phillips, 1974, 1977, 1978, 1979, 1980, 1982; Wasserman, 1984a; for a review, see Phillips, in press). This work broke a long silence on the issue that had been in effect for approximately 75 years, the period after Durkheim's polemical attack on this theory. One issue that has emerged is the effect of fictional depictions of suicide on imitative suicide (Kessler & Stipp, 1984; Phillips, 1982; Phillips, 1982). Phillips (1982) analyzed data on suicide in soap operas and found a positive relationship between such fictional suicides and imitative suicides. However, Kessler and Stipp (1984), after correcting for errors in Phillips' (1982) data, found no relationship.

While the Kessler and Stipp (1984) findings are more valid than Phillips's (1982), there is still reason for concern over rejecting the imitation hypothesis on the basis of soap opera suicides. First, all of the “suicides” on the soaps except one were suicide attempts, not completions. Hence, the audience may not have been likely to actually complete suicide in response to the suicidal behavior on the soaps. In addition, the audience for soap operas is overwhelmingly female. I (Stack, 1986c) have found that imitative suicides are largely male-dominated. Hence, we need a replication of the analysis of fictional suicides where the audience includes a substantial number of males. Films from the early 1980s that aired on prime-time television and had suicide as their theme, such as The Deer Hunter and a series of films on teenage suicide, should provide a fruitful avenue for future analysis. The final daily and monthly suicide counts for the periods covering most of these films are still not available, but should be in the months ahead. If new tests of the imitation thesis for fictional stories do not work out, it may be that stories on suicide must involve real suicides in order for them to spark imitation.

Wasserman's (1984a) reanalysis of Phillips's data, extended through 1977, found that only stories on celebrities sparked imitative suicides. Stories on plain folks, Wasserman argued, have no suggestion effect. However, Wasserman's (1984a) data were based on front-page stories in the New York Times. Many of the stories about plain folks that Wasserman used involved New York residents and did not make national news (Stack, 1986c). In addition, many of these stories were not listed in Facts on File, an index to national news; this source had been used in an earlier study of the Times (Phillips, 1974) as a means for differentiating stories that probably made national news from those that probably did not. Wasserman's study is thus in need of replication. It could be that plain folks' suicides, if they make national news, will provoke imitative suicides. In fact, a recent analysis (Stack, 1986c) found just this. Focusing on stories that made the 6:00 P.M. national televised news, I found that suicide stories on plain folks were just as apt to trigger suicides as were the stories on celebrities. Given an audience in the tens of millions on national television, suicide stories involving ordinary people can indeed trigger imitative suicides.

One question that is often lost track of in the literature on suggestion is this: How significant are media stories on suicide, relative to other, more traditional sociological variables, in predicting suicide? This issue has been increasingly neglected, given a shift to the analysis of daily suicide counts in order to pinpoint the length and timing of the imitation effect (Baron & Reiss, 1985; Bollen & Phillips, 1982; Phillips, 1982). In these analyses sociological variables are omitted, since there are no daily data on divorce, unemployment, and other conditions. Even if there were, it would seem that the impact of such factors needs to be measured over a longer time frame than a single day. Hence, designs based on daily counts of suicide have some serious methodological shortcomings, involving a rather routine error of model misspecification.

There have been a few studies involving multivariate analyses that include such sociological variables as unemployment and divorce rates; in these studies, the traditional sociological variables tend to explain considerably more of the variance in suicide than do the story variables (Stack, 1986c; Wasserman, 1984a). To the skeptic, then, it may seem that the literature on imitation and suicide may be somewhat out of proportion to the importance of the suggestion-suicide link.

I would argue that the best work on suggestion and suicide has yet to be done. Two key methodologies have yet to be applied to the data sets on suggestion and suicide. First, all of the past work is based on additive models that fail to test for interaction effects between the stories and the other variables in the analysis. For example, the impact of a story is assumed to be the same across different levels of unemployment, divorce, and other measures of suicidogenic conditions. However, it would seem likely that the imitative influence of a story would be relatively high when the suicidogenic mood of the audience was high, and low when the suicidogenic mood was low. In other words, suicide stories should have their greatest impact when the pre-existing mood of the audience is already marked by depression, anxiety, and so forth. Otherwise, stories may have little or no impact. These interactions can be modeled easily through the use of multiplicative terms in researchers' regression equations.

A second major avenue that is in substantial need for analysis is a systematic test of differential identification theory. That is, not all stories will provoke the same amount of identification between the audience and the suicide victim. After reading 300 suicide stories on the front page of the New York Times, I was struck by the enormous variation in the context of the suicide and the characteristics of the victims. The Times reports the suicides of many criminals, including bank robbers, mass murderers, embezzlers, police chiefs charged with corruption, and so on. It would seem that such stories would promote relatively little identification from the audience. In contrast, there are also stories of suicides among young lovers, people with poor mental and physical health, persons suffering from acute poverty, individuals perplexed by downward social mobility, and so on. Research is needed to differentiate stories according to their contexts and the characteristics of the victims. Only then can we fully test the vitality of the theory of suggestion and suicide. Unfortunately, there have been only a few efforts to date in this direction (e.g., Stack, 1986a, 1986c).²

While there is little micro-level research on the influence of suggestion on suicide, one often does find anecdotal evidence in support of the perspective. For example, in the instance of testing the differential-identification perspective, a recent case study supports the theory. An economically hard-pressed retired man who failed in a murder-suicide of himself and his spouse recalled the murder-suicide case of another retired man and his spouse. The latter couple were also having financial problems, and their case received widespread publicity on the television news (Gottschalk, 1986).

Another problem has been the use of limited indicators of national suicide stories. Some writers use only the print media for a sample of stories (e.g., Phillips, 1974; Wasserman, 1984a); others use only the electronic media (e.g., Bollen & Phillips, 1982; Stack, 1986c). Analysis of the overlap between these two media channels has found that there is no automatic correlation between the two. Some stories that make national television do not make the print media at all, and many make only the inside pages of major metropolitan newspapers (Stack, 1986c). Further work is needed to build an index of the overall amount of coverage in both media channels and then to assess its impact on suicide.

The Need for Comparative Work

There is a need for comparative work in order to see to what extent U.S.-based findings will be replicable in different social contexts. The variables that work in the U.S. socioeconomic-political context may not work in alternative situations.

Many comparative data are available for the cross-national researcher, both on suicide and on the relevant independent variables. Several key compilations of historical data on suicide will save the comparative analyst much time in putting together a data set. Verkko (1951) provides annual suicide counts for nine nations as follows: Finland (1890–1940), Italy (1864–1923), France (1827–1920), England and Wales (1857–1920), Prussia (1841–1861), Germany (1882–1921), Belgium (1841–1921), Ireland (1865–1911), and Spain (1883–1922). Dreyer (1959) provides data for Denmark, France, England, Sweden, and Ireland. The World Health Organization (1956) published an invaluable data set for 13 mostly European nations that goes back as far as 1901 for many nations. This data set has been updated for the 1950s and 1960s (World Health Organization, 1968). For more recent years, the researcher can dig the data out of the annual issues of the World Health Organization's (1968–1986) World Health Statistics Annual or the United Nations' Demographic Yearbook. Data on numerous independent variables are available from a few key cross-national data banks (e.g., Banks, 1971; Taylor & Jodice, 1983; World Bank, 1980) and the annual issues of such volumes as the United Nations Statistical Yearbook and the UNESCO Statistical Yearbook. Data on labor market variables are available in the International Labor Office's Yearbook of Labor Statistics, a series that goes back to the 1930s.

These data can be applied to test explanations that have been disproportionately tested with U.S. data. In all cases of the core variables discussed above in connection with the measurement of suicide, comparative work is needed because of the unique nature of the United States relative to the other industrial nations of the world. A few specific examples are offered below in hope of stimulating some comparative research inquiries.

With respect to the new RCS or underemployment explanation, I have found significant associations between RCS and the youth suicide rate for over a dozen industrial nations for the 1950–1980 period (Stack, 1984). While this certainly gives some impressive support for the Ahlburg and Schapiro (1984) model, it is also true that the amount of variation explained by the RCS factor goes from just a few percentage points to well over half of the total variation in suicide. This rather wild variation in the coefficient of the RCS term is still another sign of the need for the development of a theory of how the cultural, institutional, and other contexts of nations affect the linkage between various independent variables and the suicide rate. It is possible that in nations with traditions of strong labor movements and/or relatively high levels of political participation, RCS will elicit different responses than it does in the United States. Such responses may be more in the area of the exter-nalization of aggression. In such contexts, social movements may constitute the patterned reaction to RCS, rather than responses centered around self-destructive behavior and the associated behaviors of social withdrawal.

The religious factor may not be significantly related to suicide in some nations with different institutional structures from those of the United States. Tittle and Welch (1983) have developed and tested a theory of the impact of religion on deviant behavior that takes contextual factors into account. It needs to be applied to suicide. Basically, religion is seen to have the strongest control over individuals when other institutions that may also control behavior are relatively weak. For example, if the economy and the family institutions are failing, a strong religious institution should have a strong influence on the control of deviant behavior. In contrast, if familial and economic institutions are working well, religion may have little impact on controlling behavior, over and above the control already provided by family and economic roles.

While Tittle and Welch were concerned only with U.S. data on criminal deviance, their argument lends itself to comparative work on suicide. The U.S. is often considered the most religious country in the industrial world (Stack, 1983a). In addition, it has one of the highest unemployment rates, as indicated by the reports in the International Labor Office's Yearbook of Labor Statistics. Finally, it usually has the highest rate of divorce, as indicated by the annual data in the United Nation's Demographic Yearbook. Hence, the impact of religiosity on suicide that we observe in the United States occurs in a unique context. The strong religious institution coincides with two institutions that are relatively weak: the economy and the family. Work is needed to test the Tittle and Welch thesis for countries like Japan, which have strong economies and family structures. In such countries, religiosity may not be related to suicide.

In like manner, political variables need exploration within the contexts of other nations. The work to date on the effect of war on suicide has been based on a single nation, the United States, which has not had a war fought on its own soil since the Civil War. Marshall's (1981) thesis that war lowers suicide only through the improvement of economic conditions, then, is in need of a cross-national test. It is possible that the integrative effect of war would be present, independent of the economic effect, for countries that have had wars fought within their boundaries. The populations of such countries might indeed pull together, faced with bombardments, troop maneuvers, and so on. Having a war in one's own “backyard” may be qualitatively different from having one that is thousands of miles away. The latter experience shelters the civilian population from much of the horror that might promote integration. In addition, one might raise associated questions that have yet to be answered: For example, does defeat in war increase suicidogenic moods, and hence suicide? The answer to the latter question would also require the analysis of comparative data, since the United States has never lost a popular international war.

In the case of the impact of elections on suicide, all of this research has taken place in a single nation, the United States, which is characterized by one of the world's least politically involved populations (Taylor & Jodice, 1983). For example, the United States has the lowest voter turnout rates for elections in the industrial world. The “election death dip” hypothesis may very well work where common people get more involved in the election/political process. Comparative work on the effect of elections needs to be done in nations with high political participation.

Conclusion

The present paper has discussed methodological problems associated with core sociological variables. Due to space considerations, the discussion has been largely restricted to macrosociological work, the dominant method in the field. In all three cases of core indicators taken from economic, religious, and political life, the paper contends that comparative research strategies are needed. The United States is often an atypical industrial nation. As such, special care must be taken in accepting sociological explanations of suicide that have been tested mainly with U.S. data.

Finally, the present review is selective in the issues it raises. While it is simply beyond the scope of a single paper to discuss all methodological issues, even if the discussion is confined to macrosociology, future work should address additional important concerns. These would include the measurement of status integration (e.g. Stafford & Gibbs, 1985) and the measurement of the broad phenomenon of Durkheimian egoism-integration. In the latter case, there is evidence that we need to start combining indices of familial and religious integration, since the two are highly associated, possibly reflecting a third, common master variable (Stack, 1985). This, however, may be true only in the U.S. context.