Nonresponse bias in survey research can affect the validity of interpretation and generalizability of results (1). To achieve desirable response rates, researchers have learned to design and implement studies that maximize perceived benefits, minimize costs, and promote trust among respondents — an approach that posits survey response as the outcome of a social exchange where costs and benefits of an interaction are weighed by participants (2). Mechanisms underlying response include elements such as questionnaire design and content, participation incentives, and thank-you letters. Several studies have reported demographic differences by early versus late response status (3–6), whereas others have shown little or no differences (7,8). Inconsistencies in response-related demographic characteristics are not entirely surprising because target populations, survey designs, types of administration, and survey topic areas vary across studies.

One key predictor in survey nonresponse is the salience of the research topic to potential participants (9,10). Groves and colleagues (11) proposed a conceptual framework for survey participation that incorporates topic of interest or salience, sponsoring organization, and incentives as factors that leverage cooperation. Critical in this leveraging process is a study introduction that invokes relevance based on respondents’ personal characteristics or experiences.

The objective of this analysis was to examine survey responder status (early vs late) by sociodemographic characteristics and by salience of study variables among respondents. Because late response in an initial survey might indicate reluctance to respond to follow-up surveys, we also examined whether late responders at baseline were more likely to be late responders or nonresponders at follow-up. Understanding the characteristics of early and late responders has implications for recruitment efforts and potential response bias.

We identified 152 (8.9%) late and 1,559 early baseline survey responders. Obtaining a completed interview required a median of 30 days among late responders and 11 days among early responders. Women at high or elevated objective risk for ovarian cancer had a 73.4% (988/1,346) response rate, significantly higher (P = .04) than the 69.6% (723/1,039) response rate of women at average risk. Conversely, women at high or elevated risk were significantly less likely (P = .02) to refuse baseline survey participation (11.1% [149/1,346]) than women at average risk (14.3% [149/1,039]).

We found no significant differences between early and late responders in sociodemographic characteristics (Table 1). Among early responders, 5.2% (95% confidence interval [CI], 4.8%–5.6%) had a personal history of breast cancer compared with 3.4% (95% CI, 2.3%–5.0%) of late responders (P = .04), whereas 4.6% (95% CI, 3.5%–6.0%) of early responders had been referred for genetic counseling and only 2.0% (95% CI, 1.3%–3.2%) of late responders had been referred (P = .004) (Table 2).

We identified 90 (4.9%) late responders and 1,734 early responders in the follow-up survey. Obtaining completed interviews required a median of 30 days among late responders and 5 days among early responders. The overall response rate for the follow up was 77.9% (1,824/2,342), whereas the overall refusal rate was only 2.9% (67/2,342). We found no statistically significant differences in either the response rates (range, 77.0%–78.3%) or refusal rates (range, 2.4%–3.4%) across objective risk categories or in sociodemographic characteristics between early and late responders in the follow-up survey. However, consistent with the baseline survey, the prevalence of a personal history of breast cancer in the follow-up was significantly greater (P = .01) among early responders 5.7% (95% CI, 5.2%–6.2%) than late responders 3.0% (95% CI, 1.7%–5.2%). Because the follow-up survey did not ask about genetic counseling referral, we could not determine whether referral differences persisted by responder status.

The baseline survey/follow-up survey matched paired observations (n = 1,303 pairs) indicated that baseline survey responder status predicted follow-up survey responder status (P = .005), a result driven by the 94.0% (1,119) of baseline early responders (1,190) who were also follow-up early responders. In contrast, baseline survey late responders did not tend to repeat as follow-up survey late responders; 96.5% (109) of late responders at baseline (113) were early responders at follow-up. Nonresponse to the follow-up survey, including noncontact and refusals, did not vary by baseline survey responder status (P = .68).

In our study, a personal history of breast cancer and genetic counseling referral were more prevalent among early responders, indicating that women with experience relevant to the study research aims were more likely to respond to the baseline survey. Women at higher risk levels had higher response rates and lower refusal rates than women at average risk. Women with a personal history of breast cancer remained early responders in the follow-up survey. These findings correspond to the importance of survey salience in leveraging women’s participation.

We found no differences in sociodemographic characteristics by responder status in either the baseline or follow-up, as was reported in a survey of pharmacists (7). Although age and sex did not predict response to a survey conducted in Switzerland, lower levels of education and income and a report of poorer health predicted the need for increased reminder mailings (8). Despite a relatively high response rate, a study of health-related behaviors in the Baltic countries found late response to be weakly related to age and education (3). Thus, significant sociodemographic differences by response status indicate that demographic predictors of nonresponse are not uniform across studies (8). We would expect to find differences in demographic characteristics by response status across studies that also differ in target audiences, questionnaire content, duration of survey, and mode of administration.

We found that late responder status was generally discordant between baseline and follow-up, a potentially important finding for survey researchers making recruitment decisions on follow-up of nonresponders to longitudinal surveys. This finding indicates that additional efforts to recruit late responders to a baseline survey will not necessarily translate into similarly prolonged recruitment in follow-up surveys. That our follow-up survey refusal rates did not vary across objective risk categories suggests that perhaps respondents who had already participated in the baseline survey felt equally committed to participating in the follow up, regardless of their medical history. Compliance, or the willingness to cooperate with a request, also has been described as “commitment and consistency,” a heuristic or principle that describes the tendency to behave in a similar way in similar situations (18).

Our introductory survey materials specifically referenced our research interest in women with cancer experience, both personal and in their families. Women who themselves have experienced cancer or who have a significant family history of cancer might be most interested in a survey on “what women know and think about cancer” and “experience of cancer” because they might want an opportunity to describe their own experiences or aid cancer research. We described one of the research goals as facilitating communication between patients and physicians about cancer screening. Thus, the survey might have appeared particularly salient to women who have or have had breast cancer themselves and women who had been referred for genetic counseling. Such women are more likely to have explored their own family cancer histories, either informally or as part of the genetic counseling and BRCA1/2 testing process. In this regard, they might have felt more comfortable responding to a survey focused on family history of cancer. The same characteristics that motivate women to pursue genetic counseling and BRCA1/2 testing might also influence their likelihood to participate in related research.

This study had several strengths. First, we conducted the baseline and follow-up surveys in a large, racially diverse population that included nearly 30% African-American women, an understudied group with regard to hereditary breast and ovarian cancer. We also recruited a large sample of women at high, elevated, and average risk from a general population setting. Furthermore, our large sample facilitated a detailed exploration of the characteristics of early versus late responders in the baseline survey and a comparison of response timing between the baseline and follow-up surveys. Finally, we achieved a very good baseline survey response rate (76.3%) and retained 75.7% of the baseline respondents in the follow-up to conduct a longitudinal assessment of response propensity and timing.

One study limitation was the self-report of family cancer history. Studies have found reporting of first-degree relatives to be more accurate than that of second-degree relatives (19–21) and reporting of breast cancer to be more accurate than reporting of ovarian cancer (20,21). Thus, our objective risk classification might have underestimated the number of women at high risk, thereby attenuating differences among risk groups. Another limitation is related to the restriction of our data collection to 1 health system in the Midwest. Thus, the results might not be generalizable to other regions of the United States where attitudes about health, cancer, and personal openness might differ.

Another concern is the potential difference in data quality between early and later responders. To address these data quality concerns, we assessed the extent of missing data among demographic variables, which are asked at the end of the baseline survey, and among other salient study variables, including “don’t know” and refused responses. We found no differences between early and later responders in either the baseline or follow-up surveys. These findings provide evidence of sustained survey data quality among late responders. Our study also found no difference between early and late responders in education level, which might influence data quality. One study has shown that respondents who required a great deal of effort to attain were comparable to early responders (22). However, as we found differences in salient study variables by responder status, the pursuit of late responders in our study might have reduced response bias, at least with respect to salient study variables.

Because we do not know the exact dates of the first call attempts to responders in either the baseline or follow-up surveys, we assumed that all responders were initially contacted on the first interview day of their survey wave. Because initial call attempts occurred within the first 8 to 10 days of baseline survey wave administration and within the first 3 to 8 days of follow-up survey wave administration, we potentially misclassified some early responders as late responders. Consequently, we consider observed differences by responder status to be conservative.

The introduction of a survey topic at the critical first solicitation, either through letters or telephone scripts, is important in study recruitment. Using only generic introductory language to state survey purpose, such as to understand health in a general way, might not garner the desired response. Our findings suggest that participants are more likely to respond if the survey is of interest to them or if they believe that their responses will be deemed important. This is perhaps more relevant for special surveys than general health surveys. In addition to salience, our favorable response rate was likely bolstered by other elements shown to enhance recruitment: incentives for baseline and follow-up surveys and acknowledgment of respondents’ time and effort through thank-you letters. Finally, research also suggests that sponsorship by a legitimate authority enhances response (2).

To our knowledge, no published research has assessed longitudinal survey response patterns or potential response bias between early and later respondents in a survey involving self-reports. Our findings on establishing salience in introductory survey materials should help inform survey research and cancer or chronic disease prevention research in particular. Finally, we suggest that the extra effort made in recruiting late responders is a wise investment in time and resources.