Introduction

Longitudinal surveys are an extremely important tool in epidemiological medical research as they give clues into the natural history and causality of medical conditions [1,2]. However the outcomes of research based on longitudinal surveys are only valid if an adequate response rate has been achieved and serious non-response biases can be ruled out [3]. Thus attempts to understand what factors may predict higher response rates, in particular the role of incentives in long term follow-up studies is important.

Many of the large scale prospective national longitudinal population-based studies such as the Australian Longitudinal Study on Women’s Health have achieved good response rates over time [4-6]. For example the Australian Longitudinal Study on Women’s Health [4] reported responses rates ranging from 68% and 64% for younger, 90% and 83% for middle aged and 89% and 80% for older cohorts for surveys conducted at 3 and 6 year intervals after the initial mail out [4]. The majority of longitudinal studies like the Australian Longitudinal Study on Women’s Health however have involved repeated contact with participants over relatively regular intervals including mail out of surveys every 3 years, regular newsletters, a project website and telephoning participants to maintain contact details [7]. Clarke et al however demonstrated the feasibility of re-surveying a sample of men who resided in the United Kingdom 25 years after enrollment in the Whitehall study of London Civil Servants [8]. A random sample of 401 study survivors were mailed a request to complete a self administered questionnaire, and then asked to attend their general practice to have their blood pressure, weight, and height measured and a blood sample collected. Accurate addresses were obtained from the health authorities for 96% of the sample. Questionnaires were received from 73% and blood samples from 61% of the sample. Very little is known about potential predictors of response rates in studies that involve recontacting participants after an extended period of time.

The use of incentives may be one potential factor that could affect response rates in long term follow up surveys. While the evidence for the use of monetary incentives generally favours higher response rates [9-13], not all studies agree [14]. Koloski et al found no significant differences in overall response rates or after reminders when a lottery ticket was used compared to no incentive, but this study was cross sectional [14]. In longitudinal studies the use of incentives is suspected to be favourable although less important in terms of increasing response rates due to fact that the researchers and participants have already formed a research relationship [15].

The timing of an incentive, for example whether an incentive is sent with the survey (non contingent) or on the promise of a returned survey (contingent), may also affect response rates [16]. The timing of incentives underlies some key social exchange mechanisms. It is proposed that providing an incentive in advance of participants responding (for example with the initial mail out survey) fosters a trusting relationship between the researcher and participant, which in turn may invoke the participant to reciprocate by responding. In contrast providing a promised incentive contingent upon survey completion suggests to participants that they are being compensated for their time and effort [15,17]. According to Dillman et al, prepayments involve a participant in the research relationship which should in turn increase response rates over a promised incentive [15]. Collins et al (2000) [13] evaluated the timing of a monetary incentive to the eighth wave in a longitudinal study and found that response rates increased with prepayment versus post payment although the effect was very small compared to other cross sectional studies [9]. It is unknown what effect the timing of incentives has in longitudinal studies with minimal contact between surveys.

It is generally agreed that more females, being older and having a higher educational level is associated with higher response rates [18]. The role of other predictors of response rates to long term surveys after minimal contact is relatively unknown. Pirzada et al (2004) evaluated response rates to a questionnaire 26 years after a baseline examination with minimal interim participant contact [19]. They achieved a response rate of 59.8%. They found female sex, non-White race, older age, lower educational level, cigarette smoking, being overweight, and obesity to be significant predictors of non-response. Their sample however were older, aged 65 years at the 26 year old follow-up and were made up of employees of 84 cooperating Chicago-area organizations. Whether these results would be similar among adults of all ages from the general population is unknown.

We aimed to determine the effect of timing of incentives on response rates. We hypothesised that receiving a lottery ticket with the survey would produce greater response rates compared with a promised incentive on survey completion and no incentive. In addition we aimed to determine what baseline (sociodemographic and psychological) factors predict response rates to a 12 year follow-up survey.

Methods

Results

Discussion

This study has provided important information on the timing of an incentive and other predictors of response rates to a long-term follow-up survey with minimal contact since the original survey. Very few studies have investigated predictors of response rates to long term follow-up surveys with minimal interim contact despite the fact that these types of studies are becoming more feasible with new technology available for identifying participants up to date addresses. Longitudinal studies can offer valuable insights into disease pathology but achieving satisfactory response rates are required.

The results from the current study can reassure researchers that good response rates can be achieved in long-term follow-up studies despite maintaining minimal contact with participants. We achieved a very good response rate of 63% in this study which although is lower than achieved by some large scale longitudinal studies that have maintained regular contact with participants [4-6], was in line with response rates associated with younger cohorts enrolled in longitudinal studies [4]. At least half of our initial sample was aged 45 years and lower at baseline. Our response rate was comparable with that reported by Pirzada et al (2004) who obtained a 59% response rate to a mail out survey after 26 years of minimal contact [19]. This is despite the fact that Pirzada et al did not use reminder letters and there was no contact during the 27 years of follow-up. While the majority of this sample had not received any contact since the original survey, there was a small proportion (8%) who had participated in some smaller sub studies in the years following the initial 1997 survey but all participants had not been contacted by the researchers since 2001, a time frame of at least 8 years. We also followed the well established Dillman’s Total Design Method [15] for the follow up of non-responders. Without our replacement survey, our response was 52%. Thus including the replacement survey as well as a week 5 reminder letter resulted in the return of another 11% of surveys, which brought our overall response rate into the more acceptable range for publication [28].

While incentives are commonly used as a strategy for increasing response rates in mail surveys, the majority of studies generally favour non contingent over contingent incentives [16]. For example Gneezy and Rey-Biel (2012) et al in a cross sectional study found that compared to no payment, very small contingent payments lowered the response rate to a mail out survey while small non-contingent payments raised the response rate [16]. They support the contention that small unconditional gifts are effective in triggering reciprocity [29]. However we did not find any significant differences in terms of response rates between sending out an incentive with the survey, a promised incentive or no incentive. This extends our previous cross sectional work with the same cohort where the use of an instant scratchie lottery incentive did not significantly increase response rates over no incentive. Thus, it could be that the use of other strategies including a personalised cover letter, attractive easy to understand coloured survey, postage stamps and affiliations with the local university and hospital were more important in increasing response rates in these studies.

It is also possible that the mechanisms linked to timing of incentives are different for recruiting participants into a study for the first time compared with retaining them in a longitudinal study [15,17]. For example according to social exchange theory, offering a prepayment to a participant in the hope of creating a trusting research relationship will not be as effective in long term studies where this relationship has already been established previously [15,17]. It is important to note however that despite our lack of a significant difference in regards to the timing of an incentive and response rates we did report 5% more surveys being returned when the incentive was sent out with the survey. The decision for researchers is whether a few extra surveys are worth the cost involved with using incentives. In times where response rates to mail surveys are on the decline [30,31], particularly in longitudinal surveys, these extra surveys may make the difference between a study being publishable or a study having a larger enough sample size to work with.

Similar to previous studies [19] we did find found that being older was a significant predictor of responding to the follow-up survey. This may be due to the fact that older people are considered to be less geographically mobile than younger people and more readily contactable for follow up mail out surveys. We did not find other demographic factors that have been found to be associated with higher response rates in other longitudinal studies [19] including being female and higher educational level to be significantly different between responders and non-responders in the current study. The reasons for this discrepancy may be that those who had originally participated in the 1997 survey were slightly more likely to be female anyway and may have been more highly educated to begin with, although this latter possibility cannot be accurately determined as such data along with other demographic factors are not available for non-responders to the original survey. The survey however was carefully designed to be easy to complete with a 6^th grade reading age level so it is possible that educational level may not have been a major influence in the decision to participate in this study.

Of interest we did observe that psychological factors appeared to play a role in determining who will participate in long-term follow-up surveys. We found that people at baseline who were less neurotic were independently more likely to respond to the follow-up survey. Neuroticism is a personality trait that is characterised by persistently enduring negative emotions and over interpreting situations as threats [32] and this may be responsible for not responding to a mail out survey. Non-responders to the 12 year follow up survey also reported higher levels of anxiety, somatic distress and had poorer mental functioning on the 1997 survey although these were not independent predictors of response rates. While the relationship of psychological state and response rates in long term follow up studies is lacking, others have also confirmed that study nonparticipants have higher disease and mortality rates, poorer health status, and lower levels of functioning than study participants [33,34] and this may be particularly the case in follow-up studies such as this where participants could potentially have had these symptoms for a long period of time. These results suggest that responders to long-term surveys may be psychologically healthier than non-responders and this may be important to consider as a potential bias in studies where psychological distress is an outcome. We did not find having a functional gastrointestinal disorder at baseline to be significantly associated with response rates to the 12 year follow-up survey.

This study had several strengths. We undertook a population-based survey of 450 people from the general community. While the majority of current addresses were able to be used, the provision of a return address on the envelopes meant we should have captured the majority of return to senders and were then able to take them out of the total denominator. A small percentage (8%) had participated in earlier sub studies which may have affected the research relationship but this may have been beneficial or detrimental (participants feeling burdened by being asked to participate again). All participants however had not received any contact from the research team for the last 8 years. The results of this study however cannot be generalised to other types of incentives. While the use of an instant ‘scratchie’ lottery ticket has been shown to be effective in increasing response rates, improving the representativeness of the sample and reducing item nonresponse [35] others have shown that monetary incentives especially cash are more effective than non-monetary incentives [36,37] and should be investigated in future studies.

In summary, psychological factors may play a role in determining who responds to long term follow-up surveys although timing of incentives does not.