We used enrollment data from a large cohort study to compare the characteristics of individuals who responded differently to a mailed invitation to participate. Unlike previous studies of early response, we compared both early refusers and early consenters to late/non-responders in an effort to better understand early response of any kind. Certain characteristics were associated with early response as categorized by early refusal and early consent to participate. Multivariate regression analysis revealed that older age, higher education, White race/ethnicity, Reserve/Guard status, and working in electronic equipment repair or functional support occupations were independently and consistently associated with both early refusal and early consent.
Finding common predictors of early refusal and early consent suggest that certain characteristics may influence the probability of early response, whether explicit refusal or consent, as opposed to no response. In our study, individuals who explicitly refused to enroll required the resources to communicate with study investigators through e-mail, telephone, or written correspondence and the opportunity and motivation to use them soon after receiving the invitation. Since subjects who consented to enroll also required these resources, it is not implausible that early refusers and early consenters might share characteristics that determine or reflect the potential to respond in any manner.
Previous studies of non-response have shown higher rates of participation in mailed health surveys for women [2
], older people [5
], White race/ethnicity [7
], and for people of higher socioeconomic status [4
], although the associations are not entirely consistent. Demographic differences between early and late consenters have also been reported: subjects who require fewer mailings are more often female [10
], older [18
], more educated [1
], and White [7
]. Furthermore, studies of initial response, consent or refusal, have shown consenters to be younger, more highly educated, and more likely to be White than initial refusers (those who initially refused to participate, but agreed after recontact) [7
]. Psychological and sociological theories have been offered that explain some of these associations. For example, an application of social exchange theory posits that when an institution (such as a government or a business) administers a survey to its members, individuals of higher standing may feel the greatest obligation to contribute back, in the form of participation, to a system from which they have benefited [22
]. This could explain the higher participation rates observed in several studies among individuals of higher socioeconomic status [22
], and might partially explain why older, more-educated people, and those employed in the health care field consented more promptly in our study.
Characteristics that predict refusal are well studied. Refusers are more frequently women [23
], older [7
], non-White [20
], and of lower educational level [7
] than participants. In contrast, we found early refusal to be associated with White race/ethnicity and higher educational level. This difference may be attributable to dissimilarities in study design, as most of these studies were either telephone or in-person surveys, or it may be that certain characteristics of highly educated professional groups predispose them to explicit refusal rather than simple non-participation. For example, people of higher social standing may feel that the risk to their social position of breach of confidentiality outweighs any benefit of participation, or highly educated individuals may become frustrated more easily by multiple-choice questions that they find overly simplistic [27
]. While these theories may describe the motivations for non-participation of some of the highly educated individuals who refused in our study, they do not explain readily why this subgroup of nonparticipants chose refusal instead of non-response to express their desire not to participate or why those of White race/ethnicity were more likely to refuse. One possibility is that these individuals simply wished more strongly not to participate. Or, they may have been less timid about registering their refusal to participate with study investigators.
Access advantages may have also prompted early response. The mailed invitations included a paper survey, but also provided a Web address where an online version of the questionnaire could be completed as well as an e-mail address where invitees could request removal from the mailing list. Occupational environments that require computer skills or where email and internet access are more readily available might encourage early response using these methods. The findings from this study suggest this may be true; all occupational categories, except craft workers and service and supply handlers, were more likely to respond early. This is exemplified by significantly higher odds of both early refusal and early consent by personnel working in functional support and administration, and electronic equipment repair. Additionally, those employed in other computer-savvy occupations, such as communications and intelligence and other technical and allied specialties, were significantly more likely to consent early. Although these results suggest that access advantages may have played a role in early response, it does not explain why certain occupational groups chose refusal rather than consent.
The demographic and occupational differences found between early refusers and individuals who neither refused nor consented early may have implications for survey research methodology and epidemiologic enrollment efforts. If the characteristics that distinguish early refusers from other nonparticipants are associated with the variables under investigation, then standard methods of correction for non-participation could benefit from consideration of the heterogeneity among subgroups of nonparticipants. For example, one approach to reduce non-participation bias is to use information from a sample of nonparticipants in the statistical adjustment of results for the participants [28
]. This method might provide more precise estimates of parameters in the target population with stratification by mode of non-participation, perhaps by sampling early refusers separately from other nonparticipants.
If a subgroup of the target population is especially likely to refuse enrollment soon after being invited, the identification of this subgroup could allow costly efforts to recruit non-respondents to be targeted toward people who are ultimately more likely to enroll. In our study, many early refusers used the option of declining enrollment through e-mail. Although this required greater effort than simply ignoring the invitation, it was an easy method of refusal for some individuals who might have been less likely to refuse explicitly had refusal required written correspondence. In that case, these subjects might have accounted for an especially low-yield target for subsequent mailings. Besides reducing the cost of future mailings, eliminating early refusers from mailing lists might also prevent these individuals from feeling anger or frustration at receiving additional invitations to participate. This could reduce the chance that they would engage in organized anti-survey activity in the future [27
]. The available data suggest that providing the option of explicit refusal in a mail survey may increase the rate of explicit refusal without increasing overall non-participation [29
Although the percentage of refusers in this study was approximately 0.3%, continued contact with those who do not intend to participate can be costly, both financially and in terms of response, even if the subset is small. If, for example, repeated mailings and e-mail reminders anger refusers to the point of spreading negative press, potential responders may be swayed into nonparticipation, increasing the potential for bias. Furthermore, if participants were to refuse or consent early – that is, after the first invitation – the high monetary cost of each cycle of mailed invitations and surveys could be reallocated toward other areas of the study such as retainment.
Several limitations of this study should be considered when interpreting the results. It is possible that some of the differences in early refusal and early consent rates among subgroups of the target population are explained by differences in rates of receipt of the invitation to enroll. In this military population, younger and less-educated people may have been less likely to receive the invitation because of more frequent duty station reassignments or deployments, or lack of access to e-mail. In addition, the study population used in this investigation is a subset of military personnel and may not be representative of the US military as a whole or the general population. The US military is comprised predominantly of men and is more educated, younger, and ethnically diverse than the general US population [30
]. These differences may help to explain some of the dissimilarities in characteristics of early refusers and consenters encountered between the present study and previously published work. However, there is no evidence to suggest that members of the military have a systematically different approach to answering requests for participation, as long as participation is voluntary. Regardless, the results of this study should be interpreted cautiously, as they may be conceptually relevant to studies in other populations, but may not have similar predictors.
Although we have focused on early refusal and consent, the characteristics associated with these events may not be associated with ultimate refusal or consent. If late consenters are more similar to non-responders than early consenters are, a later comparison might show greater similarity among consenters and non-responders and greater difference between consenters and early refusers. However, the goal of our investigation has not been to identify characteristics associated with eventual consent or terminal non-response, a topic that has received much attention. Rather, the finding that early refusers share certain characteristics with early consenters, which distinguish them from those who do not respond early, suggests that this subgroup of nonparticipants may deserve special consideration in study design and analysis.
The existence of a demographically distinct group of early refusers would be less relevant to methods of response bias correction if the ultimate health outcomes under investigation were not associated with demographic characteristics. We cannot assess whether early refusers might be more or less likely to develop outcomes of interest than those who neither refused nor consented early. It may be that this question can only reliably be answered retrospectively for certain outcomes, since refusers may be at greater risk than consenters for unfavorable health outcomes, even if they are similar at baseline in demographic characteristics and general measures of health [31
A strength of this study is the availability of demographic and occupational data on all members of the invited population, regardless of participation in the study. Some studies of non-participation rely on follow-up interviews or questionnaires on a sample of nonparticipants to characterize the entire group, but the proportion of initial non-respondents who complete a follow-up questionnaire may be quite low [32
]. Follow-up interviews can be time-consuming or costly, and might only be conducted on a subset of non-respondents [2
]. Additionally, this study has a large and diverse study population, which allowed for robust comparisons between early and late/non-responders and greater generalizability than in previous studies of response.