Blog Spotlight #2: Wicked Anomie
I am quickly getting sucked into the world of sociology blogs. Fear not, legal scholars--I have not abandoned you yet. In fact, I often say to myself in my pup tent on the cold mountains of Wyoming, "I wish I could quit you, Law."
But anyway, through my marauding travels in Scatterplot, I have been discovering awesome sociology blogs. Including, WickedAnomie, who is working on her PhD in sociology, and who, like me, aspires to finish by 2010. Her blog is "sociology run amok," or "armchair adventures from the ivory tower." F'ing awesome descriptive phrases, are they not? Mine's "the chronicles of academia," but I should so steal her "armchair adventures."
For a sample of one of Wicked A's awesome posts, here is her "examining a social issue with SCIENTIFIC methods", or "do you save a cherry by popping a collar" [ed: God, I hate Abercrombie and Fitch]?:
I assert that the t-shirt manufacturers are confusing correlation with causation. They are advocating doing one action instead of the other. However, I hypothesize that doing one (popping a collar) will be causally related to a person's inability to achieve the other (popping a cherry; i.e., deflowering a virgin).
But how can I find out if this is true?
STATISTICS TO THE RESCUE!!!
Well, first we have to decide what we are predicting. This could go two ways: are we predicting the likelihood a date will end in a popped cherry, or the number of cherries popped over some period of time? I think the number of cherries popped is a better indicator of the viability of the Popped Collar Hypothesis. This way, we can examine cross-partner (or potential partner) effects.
This means that the dependent variable is a count of the number of cherries the subject has popped...[go there for actual sophisticated statistical analysis]...
It is highly unlikely that the average number of cherries popped within the general population follows a normal distribution. Hence, we cannot use ordinary least squares regression. It is far more likely that the number of cherries popped will follow the standard Poisson distribution, with a mean of about 1 (girls can only be virgins once, and I am guessing that most people only sleep with one virgin or less in their lifetime). Poisson analyses assume that the incidents (cherry pops) are completely uncorrelated, so that one cherry-popping incident has no bearing on the likelihood of future poppings. I doubt this is true. These models can't account for such an effect because there is no error term in the formula.
We can fix this with a negative binomial model, which adds an error term (dispersion parameter) with a gamma distribution. This allows the variance to exceed the mean. The negative binomial model has the same mean structure, but it now corrects for variance. The standard errors will no longer be biased downwards, hence no more inflated z-scores, and now we won't have false positives (Type I errors, where you reject a null hypothesis that happens to be true). Recall, the null hypothesis here would be that there is NO correlation between popped collars and popped cherries. This is what we are trying to REJECT.
But there's another potential problem: Poisson distributions also tend to underpredict the number of zeros. Now, if we're including boys and girls in the analysis, this is a HUGE problem. Not very many girls have popped cherries. And even fewer sport popped collars.
If we are only looking at boys, however, this may still be true because some of the zeros (boys who have never popped a cherry) are being determined by a different process than the remaining count. Some boys are being hindered by their popped collars; some are not just not going to get laid whatever they try, bless their little hearts. In this case, we can use a zero-inflated negative binomial model to test our hypothesis. What this does is (sort of) run a binary logistic regression AND a negative binomial regression on the data. So now we are kind of predicting two outcomes: who is deflowering virgins, and among those virgin deflowerers, how many virgins have they deflowered?
Then we can see if popped collars are influencing the count.
Now I just need some data.
Come on, people. Everyday sociology, studying humans with scientific methods, evaluating the sexworthiness of preppies. You must agree, the most awesome gift the the blogworld has ever given you.