Durkheim’s implausible methodology

In sociology PhD program, we are encouraged to find research interests and then decide on a research question. The methods we choose depend on what we are interested in and how we approach them. Durkheim’s Suicide does not follow this usual path. Rather, he selected suicide as a case study to “demonstrate the possibility of sociology” (37), which treats social facts as things. Given the interest in this book, this memo focuses on the methodological aspects of Suicide. Specifically, I discuss (1) Durkheim’s causality with a comparison of Weber’s one, (2) his interpretation of statistics (means and variances), and (3) his lack of interest in probability.

1.     Causality in Durkheim and Weber’s Sociology

Durkheim’s methodology is often called methodological holism (Coleman 1986). This perspective treats macro-level social things as a unit of analysis to explain other social phenomena. On the other hand, Weber is a founder of methodological individualism, in which we explain macro-level associations focusing on individual’s actions. In addition to this well-known distinction between them, another notable difference in their sociological approach is their understanding of causality. To discuss this point, it is helpful to use typologies of causality developed by John Stuart Mill. Among them, Weber’s approach is the method of difference. This approach compares two cases that are the same except for one trait. The assumption is close to the idea of a natural experiment, or “imaginary experiment” (Smelser 1976: 69). Of course, experiments are often not possible in social sciences, and some studies (including Weber’s work) look at a few historical cases. Weber’s approach is historical, not statistical, but he used this method, as we saw in the reading last week (Weber 1906=1978: 119).

In contrast, Durkeim used so-called method of concomitant variation (Smelser 1976: 63-64), which is simply a correlation between two variables (such as suicide rate and solidarity) given a variety of cases. One contrast between these two methods happens when we find third causes (Smelser 1976). On the one hand, since the method of difference focuses on the difference caused by the presence of one single cause, problems of third causes will not happen. On the other hand, the method of concomitant variation allows the possibility of third causes.

In modern statistics, we control for these third variables to examine the effect of X on Y. Durkheim attempted this without using statistics. For example, in detecting the effect of religion on suicide, he controlled for influences of culture or race by looking at different states in the same country (Germany or Switzerland). However, he was not able to distinguish the effects of education and religion, because they are strongly correlated. Instead, he found a distal cause, which is integration (159), and argued that this explains the societal differences in suicide rates. Well done.

We can be skeptical about his conclusion, however. Jewish population has a lower suicide rate (154) while their education tends to be higher (167). Durkheim picked up another arbitrary explanation, arguing that it is because of their desire for knowledge as “religious minorities”. If this hypothesis is correct, however, we can speculate that Protestants and Catholics are also highly educated when they are religious minorities in a given society. Unfortunately, this point was not examined in the Suicide. As Smelser (1976: 107) critically argued, “unsystematic appeal to a third variable was built into the logic of Durkheim’s theory.” To put it further, he was confident in his focus on integration as the explanation (Sato 2010). One methodological advantage using the method of difference is to control unobserved characteristics because they are able to compare cases with only one difference, while the method of concomitant variations is not. However, as Durkheim (1901=1982: 150) argued elsewhere, the same effect always corresponds to the same cause. In this sense, his discussion suggests that he already fixed the causal relationship before observing the data. If some anomaly happened, he added a third variable to arbitrarily solve the contradiction.

2.     Interpretation of Means and Variance

Although Durkheim criticized Quetelet’s idea of the ‘average man’[1], Hacking (1990: 178) argued that he nevertheless “stayed in the Queteletian mould,” in a sense that he still stuck with average rather than deviations (Sato 2010) or their similar assumption that “the forces acting on people were like cosmic forces or gravity” (Hacking 1990: 131). In Durkheim’s methodology, the forces are social (Hacking 1990: 177).

For Durkheim, who was interested in average rather than variation, deviations from the mean reflect pathological states of society (Hacking 1990: 178). In contrast, Galton was interested in the deviation itself. For him, the normal is just average and deviation from the mean is something we need to explain (Hacking 1990: 178, 184). This distinction has a persistent influence on quantitative reasoning in contemporary sociology (Xie 2007). Quetelet’s idea of ‘average man’ is close to what Mayr (2001) called ‘typological thinking,’ which dates back its origin to Plato. This perspective focuses on typical phenomena, and treats deviations from the mean as errors or nuisances. Thus, this perspective considers heterogeneity within the population as trivial.

In contrast, another approach towards population, ‘population thinking,’ treats deviation as a serious subject to be explained (Xie 2007). According to this perspective, a deviation is not a mismeasurement of reality. Rather, it also reflects a part of reality. Galton, who developed regression and correlation, belongs to this tradition. The latter perspective is dominant in demography, or social demography, which is interested in heterogeneity across subpopulation.[2]  

3.     Silence about Probability

As Hacking (1990: 177) pointed out, Durkheim was not interested in probability (or chance). This critique might be harsh given that he did not have individual-level data and was not able to calculate standard errors in his correlation estimates. On the other hand, however, his lack of interest in probability is understandable because, unlike survey data, the data he used targeted a whole society (vital statistics or census).

As argued, he thought that the same effect always corresponds to the same cause. Also, he theorized that suicide is not a sum of individuals but influenced by external factors. In this deterministic approach, he could justify the lack of probability in this work. That being said, it is still worth pointing out that Suicide is the book that used statistics. Weber’s approach is historical, but he does not assume these historical causes in a deterministic way (Weber 1906=1978: 116). This contrast suggests that the distinction between deterministic and probabilistic approaches is independent of which data (statistical or historical) we use.

5.     Conclusion

A summary of Durkheim’s methodology is shown below. His methodology is often compared with that of Weber, typically through an illustration of methodological holism versus individualism. Although important, this distinction masks several important criteria taken by them. Their methodology is different not only in the unit of analysis (collective versus individual), but also in causality and statistical thinking. Also, Durkheim’s quantitative reasoning is close to typological thinking, which makes contrasts with population thinking.

It is hard to discuss Durkheim’s contribution to sociological methodology (at least to me). His choice of methods of concomitant variations to identify causality results in pursuing endless arbitrary third explanations. His statistical thinking was mostly wrong. As a demographer, I take on the population thinking rather than typological thinking. We can learn more from reading Weber, especially for methodology. One contribution, however, is his focus on social forces and methodological manifests that we should treat social facts as things. This is a tentative conclusion I have so far, after reading the Rules and the Suicide.

Table Overview of Durkheim's methodology

	Durkheim	Opposing ideas
Causality	Methods of concomitant variations	Methods of difference (Mill, Weber)
Quantitative reasoning	Typological thinking (Quetelet)	Population thinking (Galton)
Statistical thinking	Lack of interest in probability (deterministic?)	Indeterministic (Weber)

References

Coleman, James S. 1986. “Social Theory, Social Research, and a Theory of Action.” American Journal of Sociology 91(6):1309–35.

Durkheim, Émile. 1901. Rules of the Sociological Method (2d ed.) (W.D. Halls trans.) Free Press. 1982.

Hacking, Ian. 1990. The Taming of Chance. Cambridge University Press.

Hauser, Philip M., and Otis D. Duncan (Eds.) 1959. The Study of Population: An Inventory and Appraisal, Chicago: University of Chicago Press.

Mayr, Ernst. 2001. “The Philosophical Foundations of Darwinism.” Proceedings of the American Philosophical Society 145(4):488–95.

Sato, Toshiki. 2010. Syakaigaku no Hoho [Methods in Sociology]. Yuhikaku. [In Japanese]

Smelser, Neil J. 1976. Comparative Methods in the Social Sciences. Prentice-Hall Inc.

Weber, Max. 1906. “The Logic of Historical Explanation.” in Max Weber: Selections in Translation. (W.G. Runciman, ed.; Eric Matthews, trans.) Cambridge University Press, 1978. 111‐134.

Xie, Yu. 2007. “Otis Dudley Duncan’s Legacy: The Demographic Approach to Quantitative Reasoning in Social Science.” Research in Social Stratification and Mobility 25(2):141–56.

---

[1] Statistically speaking, his critique on Quetelet (300-306) was incorrect (Sato 2010: 103). He first argued that Quetelet assumed that the invariability is found in some activities with which the great majority of individuals is involved, and this idea is wrong because the suicide rate, which is exceptional in population, is “even more stable than that of general mortality” (303). Those who studied an introduction to statistics know that the variability (standard deviation) depends on sample size, not the variable of interest itself. Also, the suicide rate is calculated by the number of incidences of suicide over a given population, which is exactly the same as mean, or the idea of ‘average man’. This point is also related to his silence about probability. He did not pay attention to variation.

[2] One well-known definition of demography is “the study of the size, territorial distribution, and composition of population, changes therein, and the components of such changes” (Hauser and Duncan 1959: 2).