|h/t Yard Sale of the Mind|
The Problem of Metrology
The first problem a statistician must come to grips with is that of metrology, a topic on which TOF has commented from time to time. A key question is whether the measurement system actually measures what it claims to measure. In this case, "better educated" is equated with "% of residents with college degree." But is this true?
- Not all college degrees are in real subjects, after all.
- One may be trained to proficiency in one particular field without being educated.
The Problem of GroupingFor a valid correlation, the X and the Y must be measured on the same units. This would require determining
- X: whether Voter A has a college degree
- Y: who Voter A chose
TOF is reminded of a case in which the percentage of women with cancer in a census district was correlated with the percentage of households using lawn services, drawing the inference that the chemicals used for weed control led to cancer. This was actually reported as news! Of course, there was no effort made to determine whether those suffering cancer lived in the households using the services, or perhaps next door; nor even whether the sufferer had only recently moved into the census district.
The Problem of ReificationWhen dealing with inanimate objects, correlation of % is sometimes fruitful given the prior assumption that the population drawn from is reasonably homogenous. This means that the lot or the product stream may be treated as if it were a single "individual." (Provided the process were in a state of statistical control.) But it is precisely the point that voters are not precision-made piece parts from the same stamping press. The reification of abstractions is the notion that because a bunch of people can be put in a box, they derive their opinions and preferences from the label on the box, rather than vice versa. The bucket labeled "Women" does not consist of identically distributed pieces, but of individuals with diverse opinions. Neither does the bucket labeled "College Degree." These are attributes abstracted from voters, not voters themselves. Adjectives are not nouns.
The Problem of InferenceDegree-holders comprise at best 30-40% of voters. It is possible, given the data, for every degree-holder to have chosen Romney while Obama carried the State because the poor, working class, and minorities chose him. In fact, exit polls have shown Obama's margin was much higher among HS-educated or less. This would be better in line with the mythos that Democrats are the party of the poor and working class. Higher education levels are associated with higher incomes, and higher incomes supposedly prefer Republicans.
Other possible inferences include
- Possessors of college degrees are snobs averse to those not well-spoken (Bush) or to Mormons (Romney).
- College faculties brainwash their graduates.
- College graduates are terrified of voting for a white guy running against a black guy.