Thursday, October 21, 2004
Fun with Statistics
Andrew Sullivan linked to a seemingly ironic story noting that "after steady declines under Clinton, abortion rates have been increasing under Bush." The author, Dr. Glen Harold Stassen, who claims to have been trained in statistical analysis, acknowledges that piecing the data together is difficult, but sees a marked increase in abortion rates between 2000 and 2003, in some cases double digits. He attributes this to women's economic fears, lack of a stable mate (via unemployment) and health care fears, all tied in, of course, to Democrat talking points. Now, I am not trained in statistical analysis, but have very strong mathematical skills and have performed quite a bit of statistical analysis in the course of my professional career and believe there may be another contributing factor that Dr. Stassen is ignoring.
One thing that always raises flags for me whenever someone starts spouting percentages without providing an indication of the sample population is the problem of statistical noise. Too many times I have read alarming headlines of a 25% increase or decrease in something only to find out that the initial sample population was only 4. Now, I know the abortion rate in America is greater than four, but within a stochastic model there is a minimum sample population below which predictability rapidly drops. This is called the fiducial level, and below this level of sample meaningful trends cannot be detected nor evaluated. Additionally, while data acquisition problems are noted, the potential effects of non-standard collection methodologies and inherent data errors does nothing but compound the problem of statistical noise. As Dr. Stassen notes, data collection and metrics are singularly inconsistent and lacking in the area of abortion, and while some cite privacy concerns other claim this paucity of reliable data is intentional. Regardless of why, the fact is that in many cases there simply is not sufficient data of sufficient quality to render meaningful analysis. Especially not when considering year-to-year trends and citing only two or three data points of questionable value, such as Dr. Stassen has.
Perhaps it is more revealing of my biases, but I would be much more inclined to consider a coldly impartial statistical analysis that included clear information on sampling population, data variance and discarded low quality samples. The certainty with which Dr. Stassen speaks, citing data that has been massaged into position to neatly match the exact same economic talking points Democrats have been using this election just seems a bit too contrived for me to accept whole cloth. Especially when it appears in an on-line publication that also includes two anti-Israel items, one article that describes the United States as "becoming a 'rouge nation'" that is "making the world a more dangerous place" (that also conveniently mentions the Israel-Palestinian issue as having been "stone-walled...literally"), and a rejoinder for Catholics to vote based upon "the full range of issues" because "Life Does Not End at Birth." I do not begrudge their right to advocate their positions or beliefs, but to try and act, as Andrew has, that this data instinctively passes an impartial scientific standard is flawed. It has generally been my experience that when analysis and policy are as closely tied as they are in this article it is usually to the detriment of both.
One thing that always raises flags for me whenever someone starts spouting percentages without providing an indication of the sample population is the problem of statistical noise. Too many times I have read alarming headlines of a 25% increase or decrease in something only to find out that the initial sample population was only 4. Now, I know the abortion rate in America is greater than four, but within a stochastic model there is a minimum sample population below which predictability rapidly drops. This is called the fiducial level, and below this level of sample meaningful trends cannot be detected nor evaluated. Additionally, while data acquisition problems are noted, the potential effects of non-standard collection methodologies and inherent data errors does nothing but compound the problem of statistical noise. As Dr. Stassen notes, data collection and metrics are singularly inconsistent and lacking in the area of abortion, and while some cite privacy concerns other claim this paucity of reliable data is intentional. Regardless of why, the fact is that in many cases there simply is not sufficient data of sufficient quality to render meaningful analysis. Especially not when considering year-to-year trends and citing only two or three data points of questionable value, such as Dr. Stassen has.
Perhaps it is more revealing of my biases, but I would be much more inclined to consider a coldly impartial statistical analysis that included clear information on sampling population, data variance and discarded low quality samples. The certainty with which Dr. Stassen speaks, citing data that has been massaged into position to neatly match the exact same economic talking points Democrats have been using this election just seems a bit too contrived for me to accept whole cloth. Especially when it appears in an on-line publication that also includes two anti-Israel items, one article that describes the United States as "becoming a 'rouge nation'" that is "making the world a more dangerous place" (that also conveniently mentions the Israel-Palestinian issue as having been "stone-walled...literally"), and a rejoinder for Catholics to vote based upon "the full range of issues" because "Life Does Not End at Birth." I do not begrudge their right to advocate their positions or beliefs, but to try and act, as Andrew has, that this data instinctively passes an impartial scientific standard is flawed. It has generally been my experience that when analysis and policy are as closely tied as they are in this article it is usually to the detriment of both.
