According to an article in the current Journal of Health Politics, Policy, and Law, public opinion polls on health reform are at best incomplete and at worst misleading due to a systematic bias in non-responses.  Authors Berinsky and Margolis argue that, in the context of the healthcare debate at least, non-responses (e.g., answers like “Don’t know”) are more likely to come from individuals of lower socioeconomic status, and that “these same individuals who are victims of resource inequalities are natural supporters of the welfare state and, therefore, are more likely to back health care reform.”
The authors write that, to ensure full representation of views, nonresponses should not be ignored. Instead, the analysis “should incorporate information from respondents’ answers to other questions on the survey to understand what they might have said had they answered the question.”
Imputing the views and attitudes of non-respondents is a generally acceptable method for removing bias, provided it is done carefully using suitable assumptions. Berinsky and Margolis use income as the predictive variable. According to their analysis, for example, people making less than $30,000 annually are more likely to support health reform than those making more than $100,000.
So after analyzing data from the Kaiser Family Foundation’s 2010 health reform tracking poll and imputing the views of those who did not respond, is it the case that Americans do overwhelmingly support the President’s healthcare reform after all? No. As it turns out, imputing opinions for non-responders shifts the tracking poll’s figures only slightly toward the Obama-Axelrod side of the debate—between 1 percent and 4 percent, depending on the month.
The authors acknowledge that there can be multiple causes of “don’t know” responses. For example, respondents may find it difficult to form a judgment about a challenging issue. They may find the question hard to understand, or they may regard giving an answer as too personal or too revealing in some way. Studies have shown that the behavior of the interviewer can also affect the likelihood of item nonresponse.
On matters political, might question wording outdo them all? Consider issues such as healthcare where the options presented to us feature a false dichotomy already baked in. What can an advocate of free-market medicine and limited government say to the yes/no/don’t-know question, “Does the healthcare system need reform?” Answer no and one is counted among those who defend a status quo in which nearly half of the nation’s healthcare dollars are already controlled by the government; answer yes and be counted among those who support more government controls over industry and even more spending. In other words, no for big government, and yes for bigger government. I’ve answered “don’t know” before on exactly this question, despite having a very strong opinion on the issue and a willingness to share it.
How many of these types of nonresponses do conventional answer choices generate? That might be a more interesting number to impute.
1 Berinsky, A. and M. Margolis, “Missing Voices: Polling and Health Care,” Journal of Health Politics, Policy and Law, Dec. 2011 36(6):975-987.
Jared M. Rhoads is the director of the Center for Objective Health Policy, whose mission is to advance rational, rights-respecting ideas in healthcare. The Center provides commentary and raises money to provide free books to medical students who are interested in learning about the moral and economic case for capitalism. He works as a senior research specialist in the healthcare division of a large consulting firm in Massachusetts.