Is political science a true science?

Political science. Changes in the forecast.

IS POLITICAL science a true science? Critics contend not, citing the failure of the profession to pass the acid test of a true science: the ability to predict events. But at first glance, the study of American presidential elections seems to belie that characterisation. In fact, political scientists are in basic agreement regarding what determines who wins presidential races, and several political scientists have put this consensus to the test by developing forecast models designed to predict election outcomes during the post-second-world-war era.

On the whole, these efforts have had mixed results, but a very few of them have been consistently able to predict the two-party popular vote within a percentage point or two and, not incidentally, project the actual contest winner. The 2008 election happened to be a good year for the forecast industry, with all 15 forecast models with which I am familiar, save one, predicting Barack Obama's victory. Although there were some outliers in this group, two-thirds of the prognosticators came within 1.5% of predicting Mr Obama's  final share of the two-party vote (53%), and collectively the group average hit it right on the button. Remarkably, more than half of these forecasts were issued prior to September 1st, with at least one forecast issued in January, before the final two candidates were nominated! (It predicted the Democrat nominee would defeat the Republican by roughly 51% to 49%.)

Surely that level of precision qualifies as science. Or does it? It turns out that answering that question is more difficult than one might think, as demonstrated by the release two weeks ago of Alan Abramowitz's "Time For Change" forecast of the 2012 presidential election. This is a preliminary forecast, pending the release of the second-quarter GDP number later today, but Mr Abramowitz, a political scientist at Emory University, provided several different electoral projections based on a possible range of GDP numbers (see chart).

Because most analysts are forecasting that second-quarter GDP growth will not exceed 2%, Mr Abramowitz's model predicts that the president will win less than 51% of the popular vote come November. Granted, given current national polling numbers that show Mr Obama and Mitt Romney essentially tied, Mr Abramowitz' preliminary prediction is hardly earth-shaking. Of greater interest, however, is how Mr Abramowitz arrived at this preliminary forecast.

Like most presidential forecast models, "Time for Change" has very few moving parts. In fact, in prior election years Mr Abramowitz has been able to predict the presidential race by looking at only three variables: the incumbent president's net approval rating at the end of June, the change in real GDP in the second quarter of the election year and a variable signifying whether a first-term incumbent is running for re-election, as Mr Obama is in 2012. That third variable accounts for the boost from voters Mr Abramowitz says incumbents get by virtue of already being in office.

For his 2012 forecast, however, Mr Abramowitz tweaked his model by including a fourth variable, which he labels "polarization", that is designed to account for the increasing polarisation of political discourse in America dating back to the 1996 election. He did so because during the last four presidential elections his forecast model consistently overstated the winning candidate's margin of victory (excluding third-party candidates) by 4.5% on average, as the table above indicates.

The culprit, Mr Abramowitz decided, is an increasingly divided electorate consisting of voters who are less likely to cross party lines when choosing a candidate. This means "election outcomes tend to reflect the underlying division between supporters of the two major parties, and right now that division is very close." By incorporating a polarisation variable, however, Mr Abramowitz is able to reduce his forecast of the winning margin by about half, bringing it much closer to the actual results in the last four presidential elections.

It might seem that improving the accuracy of a forecast model by incorporating a new variable is good science. But not all political scientists agree. By retrofitting his prediction model so that it more closely aligns with past results, Mr Abramowitz leaves himself vulnerable to the charge that this post-hoc tinkering lacks any substantive basis. Did the electorate really become so much more polarised beginning in 1996? Or did polarisation begin in 1992, when Bill Clinton and Ross Perot challenged the incumbent George Bush? Or perhaps it dates back to 1980, when Ronald Reagan challenged Jimmy Carter? How can we tell?

This debate over theory and model-building may strike the layperson as overly academic-as long as Mr Abramowitz nails the 2012 election while simultaneously making his historical projections more accurate, what's the fuss? But for political scientists already on the defensive regarding the merits of their profession, Mr Abramowitz's decision to retrofit his forecast model will inevitably attract more scrutiny from sceptics in the one area where political scientists hoped to make their strongest case that what they do is, indeed, science.