My rejection of the National Marriage Project’s “Before ‘I Do'”: https://familyinequality.wordpress.com/2014/08/27/my-rejecti...
I'd be interested if there has been follow up research since.
Suppose I do a study that finds lower household wealth is correlated with time spent gambling in Vegas. Do gambling losses cause the lower wealth? Or is gambling more prevalent among lower income earners?
If my goal is to choose a wealth advisor, maybe it doesn’t matter.
Vegas, regardless of 1000 other variables you might introduce, is a statistically solid system. You are likely to lose money, with the standard of deviation going down the more you gamble.
Low income, education, whatever, don’t change the strong link of gambling to losses.
And this is most definitely a causal effect. Verifiable by mathematics, any study you want to make trying testing counter vailing variables, the teleology that develops as economic systems adapt to create that causality to flourish, and the designs of gambling systems.
So of course the other variables don’t matter to your financial advisor.
—
An exception is games where skill can tip odds in your favor, however slight. Such as poker. A financial advisor isn’t going to second guess a poker player with a long standing career of positive outcomes.
I.e. if A is a reliable predictor of B, it doesn't matter whether A caused B or both A and B are caused by some other factor; A is still a reliable predictor regardless.
it is funny, really how obvious the angle is in these kinds of analysis. There's no way you could even start to draw conclusions from these kinds of data sets unless you already knew what you were going to report -- there's less than no information in the numbers being used... the only rational analysis would result in nothing but questions and absolutely no conclusions.
(RIP Bill Hicks)
>The method for arriving at these “adjusted” (for race, education, etc) percentages is not given.
There is a pretty standard way of doing this: you stratify the sample, then re-weight the subcategories according to their distribution in the population. For example, if your sample had 10% Black people and 15% Hispanic, you would re-weight to 12% and 18%. There are valid criticisms, and then there is objecting for the sake of objecting.
Meanwhile, the original study has its own share of eye-spinners:
>Many in Generation YOLO (you only live once) believe that what happens while you’re young won’t affect your future.
Generation "YOLO" drinks less as teenagers than any generation before them since Prohibition. Generation "YOLO" has less sex in high school.
>There are two other related premarital risk factors for low marital quality in our research sample: Having lived with someone other than a future spouse and/or having been married previously.
The effect shown in the figure is very small. As a rule of thumb, the smaller an effect is, the more likely it can be caused by hidden variables. Smoking causing a 1600% increase in lung cancer rates was probably not hidden variables. A 20% increase in the frequency of "high-quality" marriages could be caused by almost anything.
Following the same logic, there are a few inferences drawn in the study that are likely to be robust: among college-educated adults (but not non-college adults), those who had a child together before marriage were fifteen times less likely to report high marriage quality (Figure 5). Even if we object to the way data were collected and binned, it's hard to put this rabbit back in the hat: it probably means something, even if it isn't causal.
The major weakness of the project is its grandiosity: its assertion that it has identified all sorts of lifestyle factors that can probably lead to a high-quality marriage, and that it can somehow serve as important life advice on these grounds. This drives an excessive optimism about conclusions that, mostly, should be checked again several times before being accepted.