Posted by Solomon on 8/12/2015, 5:21 am, in reply to "NOIR OF THE WEEK 8/11/15: CASABLANCA (1943) meets the NoiroMeter"
68.133.16.234
Outstanding work, Don. It's useful to have judgments of noirness presented in an explicit way. A few observations on this approach.
In a statistical approach to the problem of measuring noirness, the dependent variable (noirness) would have some objective measure. So would the independent variables. The analysis would deliver the weights that produced the lowest errors in the sum of squares. You've therefore done a kind of multiple factor ANOVA here without using the explicit statistical techniques that achieve the greatest explanatory power or least error of classification. ANOVA is analysis of variance. You have a kind of discriminant function here. But there is a big difference. We do not have these natural or objective measures in this case, so we are involved in a kind of circularity that makes backtesting essential. Within the scheme and before applying it to a specific film, the percentage weights are judgmental as are the values attached to each. However, both the factors and weights had to be chosen so as to reflect films that were already identified as noirs. They had to be chosen so as to deliver a rating that classified such noirs according to predominant existing identifications of noirness after step 2, the judgments of scores, was completed. There had to be backtesting on "known" noirness in films. I'm aware that you've done this on many wellknown films. I'm only pointing out that this was and is an essential part of the process of verifying that the model is actually measuring the noirness that it sets out to measure. At the same time, when the model is applied to a new film that's out of sample, it's measuring how closely such a film conforms to those films from which the meter was devised in the first place. There are some cautions to be observed in applying this method. The factors may not all be independent. As time passes and films change, the nature of noirness may change too. New factors may emerge, others fade. The relative weights may change. Gambling may be a factor in one era but not in another. A new symbol for some noir factor may come into play or a new factor altogether. The essential problem of there being no natural measure of noirness is the root problem. This is alleviated by defining a classic noir period, possibly itself subdivided, and distinguishing it from postnoirs and neonoirs. These periods proxy for new kinds of noirness or new weights while maintaining some affiliations of the different periods. 
Responses
