After reading the Feb. 13 article by Joe Adgie entitled “Student majority opposes ban,” I am left with significant concerns regarding the accuracy of the reporting it contained.
In the front-page piece, Adgie notes that a majority of students−53 percent−opposes a proposed Board of Regents smoking ban, based upon a Web survey promulgated via email by the Student Government Association. In a previous article, Adgie notes as of Feb. 10, there were 486 respondents to the SGA survey with 42 percent against the policy.
Let’s assume there are now approximately 500 respondents with an aggregate student population exceeding 12,000−a mere 4 percent responded to the SGA survey (that’s not including faculty and staff). Any student or faculty member in a department which employs statistics in their research will likely concur that the Web-based SGA survey is methodologically flawed.
This survey is likely a classic case of self-selection sampling bias, a statistical issue where survey respondents are allowed to decide for themselves if they want to participate in the survey or not.
Thus, in this case those most adversely affected by a potential USG smoking ban or who feel most passionately about the ban are most likely to participate in a survey which seeks their input. The result, as this survey reflects, is a low number of respondents with a close division between groups.
There are ways to correct for selection bias (most notably, Heckman’s lambda), though I doubt the SGA employed these techniques in their survey. We often say in political science, the media−including The Spectator−have an agenda-setting role; they don’t tell us what to think, but what to think about.
It is irresponsible of a news outlet to assert, definitively, a perceived majority opinion of a campus community based upon a flawed e-survey. It would be more appropriate to say, “Of those surveyed, a majority are opposed to a smoking ban.”
There’s another statistical term for a researcher who posits too broad of a conclusion based on their data−it’s called an overgeneralization. And that’s what I think is happening in this case.