Participation rates in HUNT3 depended on age, sex, socioeconomic status and type of symptoms and diseases. Among nonparticipants, the prevalence of common chronic diseases was higher compared to that reported by participants. This included cardiovascular diseases and diabetes mellitus, a pattern confirmed by prevalence data based on diagnosis by GPs. Contrary to this, at least among people younger than 80 years, common problems like musculoskeletal pain, urine incontinence and headache were reported more often in participants compared to nonparticipants. The study confirms associations between participation and marital and socioeconomic status, and maintenance of increased risk of death for nonparticipants even many years after the surveys. Combining data from different sources provides the opportunity for future sensitivity analyses of prevalence, incidence and association studies.
Except for the Tromsø Study , most national [3, 5, 6, 15] and international studies [1, 16, 17] have reported corresponding reduction in participation rate as the HUNT study. Reasons for the increase of nonparticipation in population based studies are thoroughly discussed by Galea et al. , and may also be relevant to the Norwegian population; People might be overloaded with invitations to research and marketing surveys, there is a general decrease in volunteerism parallel to decrease in participation in organizations and social activities, there might be lack of immediate benefit for the individual participant and, generally, there is an increasing disillusionment with science due to conflicting results between different studies and changing recommendations for behavioral risk factors. Further, more complex procedures regarding informed consent and study protocol as well as the burden of being invited to follow-up studies, might decrease study participation [18, 19].
The HUNT surveys have also become more complex and demanding for participants with more comprehensive questionnaires, interviews and examinations. Further, in order to keep to laws and regulations for medical research, an eight pages information folder was sent together with the personal invitation and consent form, even though invited persons prefer more simple forms . Increased number and size of follow-up questionnaires (Q2 and Q3s), however, have not influenced a stable response rate of these of about 75-80% among those having attended the examination stations in three surveys. This indicates that the length and number of questionnaires have minor effect on the all over participation in those who from the start had decided to participate [20–22]. Our data show higher participation rate in HUNT3 among persons previously having participated in HUNT2 compared to nonparticipants, this is in line with results from another Norwegian Cohort Study . Previous participation status in HUNT2 did, however, not influence the participation rate for QNP, indicating similar attitude to contribute in data collection by short questionnaires in participants and nonparticipants.
Low participation rates among persons under the age of 40 probably reflects less opportunity to spend time, limited short time benefit for a rather healthy age group [1, 24] and less altruistic attitude in contributing to research . Higher prevalence of chronic diseases having regular follow-up by health care among nonparticipants and opposite pattern for bothersome, but more trivial symptoms in HUNT3, do support potential benefit for the individual person to be of importance when considering participation. With increasing diagnostic and therapeutic opportunities in the health care during the last decades, present surveys are considered less as a supplement to the health care.
The present study confirms previous studies having reported low participation among people who are young, unmarried, and belonging to lower socioeconomic groups [5, 6, 21, 25–28]. Interestingly, also in the HUNT Study persons in higher socioeconomic groups, with presumably highest time pressure, participated more often compared with lower socioeconomic groups, indicating that motivation and attitude to research are important.
About 10% of nonparticipants claimed that they had not received the invitation. We have no indication that this can be explained by failure in the post delivery. However, the invitations were sent in plastic foils, and this could have been taken as advertising leaflets and therefore not read. Economic costs should not have much influence on the participation rate; the participation was free of charge and most people were allowed by employers to meet during working time with full salary. To avoid nonparticipation due to the inconvenience of leaving work, the examination stations were open even in the early evenings. Participation was not stimulated by financial incentives, as previous studies have not found this to increase response rates .
There has been a general positive attitude towards health related research in the population of Nord-Trøndelag. Since HUNT2 (1995-97) there have been regular reports from HUNT-related research on the HUNT web-site and in the media. The HUNT Study has been supported by the Norwegian Parliament, the Government, the Ministry of Health, and there has been a close collaboration between the County Council, the municipalities and HUNT Research Centre. Additionally, the media have been strongly supportive of the HUNT Study.
Influence on incidence and prevalence estimates
Most studies have found little evidence for substantial bias due to nonparticipation [6, 20, 28–30]. In a Norwegian community respiratory cohort study, increasing the response rate from 65 to 89% after three reminders, resulted in no overt differences in incidence rates of respiratory symptoms and asthma as well as their associations to sex, age and smoking habits . However, like the present study, others have reported underestimation of psychiatric disorders due to nonparticipation [5, 32, 33]. The scores of HADS anxiety and depression symptoms cannot be directly compared with GP’s diagnosis of anxiety and depression. However, our finding of higher prevalence of reported anxiety symptoms compared to anxiety diagnosis, and lower prevalence of reported depressive symptoms compared to depression diagnosis, in participants compared to the background population, indicates that depression is a more restricting factor for participation than anxiety.
In line with other studies, nonparticipants in HUNT3 was also characterized by more unhealthy lifestyle regarding tobacco smoking and physical inactivity [26, 34], and poorer somatic status [25, 30, 35, 36]. Data from the GPs further indicates as much as 50% higher prevalence of angina, myocardial infarction and stroke compared to the HUNT3 data, a pattern also found in other cardiovascular studies [25, 35]. Correspondingly, the diabetes prevalence based on HUNT3 Q1 was underestimated both compared to QNP and the GP data based on diagnosis and prescription of medication. Higher prevalence indicated by GP diagnosis than by prescription data could be explained by prescription of insulin by hospital doctors for young adults, and attainment of adequate diabetic control by change in life style for many patients with diabetes type II.
The difference in BMI estimated by self-reported measures compared to measures at the examination stations in HUNT3 is in line with data from 2008 in an Australian study . Lower participation rate among lower socioeconomic groups could contribute to reduce the difference between self-reported and measured anthropometrics.
Influence on associations
Many studies have found that subjects with risk behavior like smoking, high alcohol consumption or drug use are underrepresented in studies addressing these factors [38, 39]. The corresponding pattern in the present study is mainly supposed to be related to differences in participation due to socioeconomic factors, as life style factors by themselves amounted to a small fraction of all questions. Studies on effects of environmental and occupational exposures have experienced participation bias depending on exposures measured [40, 41]. By analyzing exposures in blood samples and linking HUNT data to external registers, a lot of associations can be analyzed, but these should not be influenced by participation as these topics have not been focused prior to the surveys.
Nonparticipation bias may be greater in surveys with higher participation rates compared with those with lower participation rates , as the differences between participants and nonparticipants may exaggerate real differences between participants and the eligible population sampled . In the present study, however, inclusion of nonparticipants answering QNP increased the prevalence estimates for chronic diseases slightly, but this correction decreased the gap slightly in prevalence between the record data from general practice and HUNT participants. If we assume that the prevalences found among QNP participants 60 years and older, were representative of the entire nonparticipant group, the combined prevalences for the nonparticipants and Q1 participants would be close to the estimates found in the GP population for myocardial infarction and asthma, whilst for COPD, apoplexia and diabetes the GP population would still have higher prevalences. Strategies for improving participation rate should be developed and evaluated prior to the next HUNT survey. In this population increased participation rate seem to improve the external validity of prevalence estimates. After HUNT3 there has been a focus on overweight, obesity and laziness in the media. This might introduce differential influence on participation in future surveys due to stigmatization or victim blaming.
Nonparticipation has been found to be associated with two times higher risk of death independent of socioeconomic category in nonparticipants compared to participants, both in a 20 years follow-up study of white collar workers aged 35-55 years  and the FINRISK study, inviting persons aged 35-74 years . Higher mortality risk among nonparticipants could be due to disabling diseases hindering study participation. To avoid inclusion of end stage patients in the present study, start of follow-up after each of the HUNT surveys were set to 2 (or two) months after study inclusion.
Strengths of this study were inclusion of several data sources for evaluation of nonparticipation bias, GP data representing background data regarding diagnoses and use of medication for the entire population, register data revealing potential nonparticipation bias by socioeconomic status and QNP reflecting the nonparticipation group. The representativeness of responders to QNP for the entire nonparticipation group might be questioned, but sensitivity tests for chronic diseases indicate that the validity of these data is rather good.
Limitations of the study were inability to link data from general practices to HUNT-data due to patient confidentiality, use of registered diagnoses in general practices that could lead to some extent of underestimation of prevalence if doctors had not registered diagnoses given by previous GPs, and comparison of chronic diagnoses with symptom report for the last 12 months (for example head ache, urine incontinence). Background data from the GPs were collected three years after HUNT, but this should not introduce important bias in prevalence estimates.
Multiple testing may contribute to statistical significances by chance, but having this in mind, there seem to be consistent results on differences that researchers have to take into consideration.