Yihui Xie

Ludwig von Mises; Conditional Logistic Regression

谢益辉 / 2007-05-01

Today I continued to help Dongxu CHENG with her SAS installation. Usually I call SAS “Stupid Analysis System” and SPSS “Stupid Package for Social Science”; though I’m kidding, the fact is that they are sometimes really stupid… Here I just want to remind researchers of statistics a fact that ASCII data are always convenient for any software. Don’t rely on Excel! Save your data in ASCII format (such as *.txt, *.csv, etc) before importing. Yes, it’s not the same with databases. (In such cases, use special database language instead, e.g. SQL)

This evening I thumbed through a part of Ludwig von Mises’s book Liberalismus (German edition, 1927, and English translation as “The Free and Prosperous Commonwealth”, 1962). Suddenly I felt that we should talk something about such economists in our CPC courses because some of their views are very practical and they will help us understand (meditate on) socialism further. Below are two paragraphs in the end of his book (I believe Mises is a great writer; readers can be easily affected by these words using parallelism):

“Liberalism is no religion, no world view, no party of special interests. It is no religion because it demands neither faith nor devotion, because there is nothing mystical about it, and because it has no dogmas. It is no world view because it does not try to explain the cosmos and because it says nothing and does not seek to say anything about the meaning and purpose of human existence. It is no party of special interests because it does not provide or seek to provide any special advantage whatsoever to any individual or any group. It is something entirely different. It is an ideology, a doctrine of the mutual relationship among the members of society and, at the same time, the application of this doctrine to the conduct of men in actual society. It promises nothing that exceeds what can be accomplished in society and through society. It seeks to give men only one thing, the peaceful, undisturbed development of material well-being for all, in order thereby to shield them from the external causes of pain and suffering as far as it lies within the power of social institutions to do so at all. To diminish suffering, to increase happiness: that is its aim.

No sect and no political party has believed that it could afford to forgo advancing its cause by appealing to men’s senses. Rhetorical bombast, music and song resound, banners wave, flowers and colors serve as symbols, and the leaders seek to attach their followers to their own person. Liberalism has nothing to do with all this. It has no party flower and no party color, no party song and no party idols, no symbols and no slogans. It has the substance and the arguments. These must lead it to victory.”

This book can be downloaded in PDF or read online. It has also been translated into Chinese in 1994.

Conditional logitstic regression surely has nothing to do with Mises; that’s just what I studied this evening. The idea is based on the conditional probability of $n_i$ “positive” ones in $N_i$ cases in the $i$-th group. Usually we don’t care which $n_i$ cases are “positive”, but in conditional logistic regression, we do care – that’s why conditional probabilities arises: P{these ni cases are positive} / P{ni cases are positive}. Some relavent terms are “Risk Ratio” (RR), “Odds Ratio” (OR), and “Relative Risk Ratio” (RRR). The last one appears in multinomial logistic regression.

And I also read a paper on instrument variables (IV) and general method of moments (GMM), but gave up at last because they were a little bit too difficult for me. My knowledge of mathematical statistics is still insufficient.