Polls, the Election, and Public Health Research: Reaching the Hard to Reach

With the election of Donald Trump as the 45th president of the United States, speculations about the impact of his presidency on the nation’s health care policy have begun. A first lesson, however, can already be learned. Over the past year, polls continually suggested a Democratic president, but ultimately the Republican candidate was elected. Surely, this will be the subject of in-depth research, and several explanations will be considered. For example, survey respondents may have changed their minds in the period between the polls and the elections as a result of new information. The reopening of the Federal Bureau of Investigation’s investigation into Hillary Clinton’s mail server use is an example of this mechanism. This cannot be the entire explanation, though, as the polls still predicted her to win the day before the election.

An often heard, but more difficult to investigate, explanation relates to a potential discrepancy between expressed intentions and actual voting behavior. Perhaps more than in previous elections, social desirability bias is part of the explanation. Also, pollsters had to deal with undecided voters, which may have given polls an almost unavoidable uncertainty. Furthermore, numbers are easier to translate into percentages than into the distribution of electors, and this is why polls did a better job of predicting the popular vote than of predicting the majority of electors.

A crucial element of further study and discussion, however, may relate to the turnout of groups that either could not be reached by the polls or were not willing to participate in the surveys. Indeed, the average voter turnout of almost 56%, with a highest rate of more than 68% in New Hampshire, is substantially larger than the approximately 10% response in many polls.^1,2 Although a low response is not necessarily a problem in itself, selective participation easily gives rise to survey error. An analysis of the 2015 UK general elections suggests that of all these reasons, low response in surveys leading to unrepresentative samples may be the most important problem.³

ELECTION POLLS VS PUBLIC HEALTH SURVEYS

This finding relates to what is observed in public health research. Those not participating in election polls may also not be inclined to participate in public health surveys. Indeed, survey research is increasingly grounded on low and selective response rates. Involving so-called hard-to-reach groups in public health research is complex and is particularly relevant in health disparities research.⁴ To describe the magnitude of health disparities by education or income levels, the participation of substantially large random samples of each socioeconomic stratum is pivotal.

What would the consequence be if everyone who voted also completed health surveys? The possibility cannot be excluded that health disparities would be found to be larger than is currently reported. Regardless of the direction of the impact, however, the polls remind us that reaching and including groups that are currently underrepresented in studies may affect the final result and therefore deserve more attention.

Although this point is by no means new,⁵ if we are serious about accurately describing the size of health disparities, more emphasis on selective nonresponse is warranted. For example, research suggests that poststratification is not always sufficient to remove the impact of selection bias in surveys entirely, and the application of advanced methodology needs serious consideration.⁶

NONRANDOM SELECTION

What adds to the complexity of predicting elections is that voter turnout often is not a random sample either. In fact, presidential candidates are very much aware of the importance of motivating subgroups in the population to cast ballots. In the end, these groups can make a difference.

Actual participation in public health interventions is also not random. An equity-specific analysis of obesity-related lifestyle programs, for example, illustrates very low participation rates among the lowest socioeconomic groups.⁷ Again, such findings raise the question of how to engage these groups in research and intervention studies.

REACHING THE HARD TO REACH

Clinton’s lack of campaigning and loss in Wisconsin suggest that reaching population subgroups is a matter of effort, but this is probably only a small part of the story. There is a real need in public health to better understand the reasons for the low participation rates of socioeconomically disadvantaged groups in surveys and in public health interventions. As in the decision of whether to vote, the broader circumstances in which persons live and work may play an essential role.

For example, those who experience financial problems, poor housing circumstances, and work-related stress may give participation in research or interventions a much lower priority than those who do not. Approaches need to be developed that align with these circumstances. A first lesson from the elections is that reaching hard-to-reach groups is a currently underestimated issue that deserves a more prominent place in public health research.