偶像的黄昏？定量社会学者说叶启政的“均值人”理论过时了？

华语社会学理论前辈叶启政于2018年出版了《实证的迷思》，延续了他对实证主义的批评。这几天，微信朋友圈都在转这本著作出版资讯。

该书目录见：

序 渠敬东

自序 

第一章 经验实征取向主导下的美国社会学 

第二章 社会现象的量化、测量与数学化 ——向自然科学倾斜的认知典范 

第三章 均值人与离散人的观念巴别塔 ——统计社会学思考迷思的两个基石 

第四章 再论“均值人”的社会学意涵 ——“机遇”与“命运”的天敌 

第五章 社会学家作为说故事者

其中，“均值人”仍是重要关键词。第三章原发表在《台湾社会学》2001年第1期，也被收录于2005年于群学出版社出版的叶启政文集《观念的巴别塔：当代社会学迷思》。

但是，华语社会学界，一向少见定量社会学者对叶启政老师的评论，现刊出两份：一份是现任香港中文大学社会学系、原“中央”研究院社会学所欧美所谭康荣教授。另一份是香港理工大学应用社会科学系首席讲师荣休吕炳强。

一、谭康荣，2005，《Crushing the Berlin Wall of Sociology: A Review Essay on The Babel Tower of Ideas》，台湾社会学，第9期。（该文是以英文发表，对叶启政老师的《观念的巴别塔》一书的书评。

谭康荣总结的叶启政观点

The assumption is that the twin ideas of central tendency and dispersion have been the cornerstones of statistical sociology since the early nineteenth century. The central claim is that contemporary statistical sociology is inevitably trapped by the iron cage of its historical origin in general and Quetelet’s interpretations of statistical concepts in particular.

谭康荣的观点

The book also has a disappointing tendency to rely on slippery deduction and undocumented empirical conjectures. ... The presumed foundational significance of central tendency and dispersion for contemporary statistics is hugely overstated.

具体的批判

1. The two statistical ideas的使用起来一定反映为“均值人”吗？

Moreover, the most frequent use of central tendency and dispersion by quantitative studies in professional journals involves model estimation rather than descriptive statistics: the biasedness of a point estimator (whether its central tendency is different from the true parameter) for an effect parameter and its standard error (the dispersion or statistical uncertainty with which a point estimate is obtained from a sample). The context and application does not mirror the notion of the average man and dispersion from it, as an effect parameter is not about properties of individuals but about a conceptual model, hence far removed from what Quetelet had in mind.

2. 把19世纪的Quetelet概念的影响力扩展到当代，却没给出关于这样当代影响力的证据，在研究设计上有问题

The book does not present any evidence for the continuing influence of Quetelet’s conception of the average man on contemporary sociologists ... I have argued that the role of central tendency and dispersion is mainly in descriptive statistics and so their role in quantitative sociology is minimal. ... The research design of the chapter is to focus on Quetelet of the nineteenth century and completely omit any reference to contemporary quantitative sociologists and their practices.

3. 因果性的当代发展，如Pearl 2000

Skepticism about the possibility of empirically identifying causal effects turns out to be the result of muddled thinking about correlation and causality. In a path-breaking treatise on the subject, Pearl (2000) presents an ingenious synthesis of the cumulative wisdom of artificial intelligence, philosophy of science, statistics, and econometrics, and a seminal graph-theoretic approach to bridge practical intuitions and rigorous mathematical proofs. Although confusions and doubts have sustained for decades, by now it is clear that causal ideas, models, and conjectures can have precise representations, easy-to-understand practical rules of identification, and so are often testable even without strong parametric assumptions of functional form or distributional forms of observed and unobserved concepts

4. 类比推理地反驳

Indeed, the two statistical ideas are the most common in what elementary textbooks classify as descriptive statistics.

Let us consider an analogy with English: the alphabet “a” (as in “a”, “are”, and “and”) is likely the most common alphabet that appears in English writing, yet the alphabet “a” is no more foundational than any other letter of the alphabet. Most importantly, central tendency and dispersion per se, do not tell us anything about foundational statistical principles, just as letters of the alphabet tell us nothing about the grammatical and interpretive principles of English. In short, the two ideas might have been quite central to descriptive statistics and to statistical thoughts of the nineteenth century, yet they are far from representing the foundations of contemporary statistical sociology.

5. 有比 the two statistical ideas更重要的统计原则

In fact, as far as statistical principles are concerned, more fundamental than central tendency and dispersion are factors such as (1) the distinction between a population of heterogeneous elements and a sample from the population, (2) variation across samples, (3) the estimator and the sampling distribution of an estimator, (4) various forms of the law of large numbers,5 and (5) models of statistical relationships.

二、吕炳强

叶老师一书，从目录看，知识范围似乎是停留在英国皇家统计学会（Royal Statistical Society）官方的GLIM（Generalized Linear Interactive Modeling) 软件之前流行的统计学方法。GLIM 是我上世纪80年代后半期至90年代做实证研究的至爱，是现在流行的 R 这统计学软件包的前身。因此，对于90后出生的一代社会学研究生来说，叶老师这本书确实会有点过时的味道。

对于我来说，GLIM 的深幽在于：在严格的概率理论（probability theory）基础上和实证研究无法避免的测量理论（measurement theory）限制下，整个假设—推导主义（hypothetico-deductivism）却得到近乎解放的寛容。也就是说，实论研究者几乎完全可以在社会学理论（即在某些条件下能承受实证探究的社会理论，实证研究者就是社会学理论家）里自由想象。

正是在这样自由的社会学理论想象之中，才有可能设想理论社会学（抢当库恩（Thomas Kuhn）意义上的范式（paradigm）的社会理论）从刚硬的数学系统（mathematical system）转变为柔软的语意学系统（semiotic system）。不知叶老师是否考虑过这些？

（Sociological理论大缸第191期）

链接：