Factors Affecting Response to Mail Questionnaire : Research Topics, Questionnaire Length, and Non-response Bias

To evaluate the feasibility of mail questionnaire for eidemiological studies in general population in Japan, we mailed six types of questionnaires which differed in lengths and topics to male residents drawn from a telephone directory of Gifu City in Japan. Those who were allotted to the shortest questionnaire and not responded after the initial mailing were assessed by followup processes with a second mail and a personal contact by telephone. We classified the subjects as early (on the first mailing) and late (on the second mailing) respondents, and nonrespondents (respondents to telephone interview), and compared histories of cigarette smoking and drinking, and environmental exposure to solvents among these three groups. The response rates after the first mailing ranged from 32-54% across the six types of questionnaires. Although somewhat lower response rates were noted for the longer questionnaires, there were no significant differences in the response rates by type of questionnaire. Although the second mailing increased the response rate to nearly 70%, there was a evidence of non-response bias. Cigarette smokers and drinkers were more likely to be the nonrespondents. J Epidemiol, 1995; 5 : 81-85.

To evaluate the feasibility of mail questionnaire for eidemiological studies in general population in Japan, we mailed six types of questionnaires which differed in lengths and topics to male residents drawn from a telephone directory of Gifu City in Japan.
Those who were allotted to the shortest questionnaire and not responded after the initial mailing were assessed by followup processes with a second mail and a personal contact by telephone.We classified the subjects as early (on the first mailing) and late (on the second mailing) respondents, and nonrespondents (respondents to telephone interview), and compared histories of cigarette smoking and drinking, and environmental exposure to solvents among these three groups.The response rates after the first mailing ranged from 32-54% across the six types of questionnaires.
Although somewhat lower response rates were noted for the longer questionnaires, there were no significant differences in the response rates by type of questionnaire.
Although the second mailing increased the response rate to nearly 70%, there was a evidence of non-response bias.Cigarette smokers and drinkers were more likely to be the nonrespondents.J Epidemiol, 1995; 5 : 81-85.
mail survey, response rate, non-response bias, questionnaire length, research topics Mail surveys are widely used to collect data for epidemiological studies.The major advantages of use of mail questionnaires are a lower cost and effectiveness to obtain a large amount of information from widely dispersed sample simultaneously.However, validity of mail surveys is limited by the response rate, which tends to be lower than personal interviews.With a low response rate, bias arising from non-response might distort the findings.
The efforts to raise the response rate have been made and some very high response rates from general population have been achieved1-3).However, it is generally believed that mail survey to the general population in Japan will produce a low response rate because Japanese people may be reluctant to diverge personal information.Kono et al4).searched for control candidates to a population-based case-control study of acute myocardial infarction by using the telephone directory of a city in Japan.Out of 1,068 persons randomly selected from the directory, 61.8% notified whether they were eligible for controls or not after three mail inquires.Hoshiyama et al.5) sent the letters requesting participation in the lifestyle survey to 4,052 residents randomly selected from the electoral roll.The participation rate was 27.5% without reminders.Although a number of studies have utilized the mail questionnaires, most of them have been applied to the particular groups from which the investigators would expect the high response rates, such as patients at hospitals, students, and certain occupational groups.
The purpose of the present study is to evaluate the effectiveness of mail survey to the general population in Japan.We examined the effects of two factors, i.e., length and topics of questionnaire, on the response rate.Although many researches showed that questionnaire length was not related to response rates 6-9) most of them achieved high response rates.The length may be important in the population from which we expect low response rate.We compared the response rates of six types of questionnaires which differed in topics and lengths.Non-response bias was measured by follow-up processes with a replacement questionnaire and a personal contact by telephone.

MATERIALS AND METHODS
The study subjects were recruited from male owners of telephones in Gifu City which is located in central parts of Japan with population of about 400,000.A telephone directory published in 1994 was utilized to draw a sample.Assuming the numbers of rows of owners' names listed in the telephone directory are approximately equal in all colums at all pages, we alloted the consecutive numbers to the rows, colums and pages.Potential subjects were selected according to the row, colum, and page numbers encountered in the random digits table.If the digits corresponded the non-individual name or apparent female name, the next digits was examined.
We provided six forms of questionnaires.Topics included in each form of the questionnaires are listed in table 1. Form 1 was the shortest and contained questions common to all forms.Form 6 included the same questions of Form 5, but they were crowded into two pages.
The questions about cigarette smoking comprised the numbers of cigarettes smoked per day and years of smoking, and ages when started and quitted smoking.Alcohol consumption was assessed by type of liquor, duration and frequency of drinking, and usual serving size.The past medical history included questions about a series of six chronic and infectious diseases (mumps, rubella, rheumatism, asthma, lymphoma, and cancer), and blood transfusion.The age at which the disease was first diagnosed was asked.For family medical histories, five hematopoietic disorders (leukemia, lymphoma, myeloma, aplastic anemia, and refractory anemia) were selected and respondents were asked whether the first degree of relatives had ever had such diagnoses.All questions were either called for circling a code number or writing a number.The reason why we included these blood disease related items was that we also need to estimate the response rate before conducting a population-based case-control study of a hematopoietic disorder.
We alloted 600 study subjects randomly to the six forms and sent the questionnaire to each of them, along with cover letter asking his participation in this survey and a self-addressed, stamped return envelope.With this assignment, approxmately 20% of difference in response rates can be detected with the significance level at 0.05 and the power at 0.80.If the mails were returned being notified that the subjects were not localized, next feasible persons were selected from the telephone directory.The proportion of them were less than 3 percent of the original samples for any form of the questionnaires.After 18 days of the mailing date, a reminder with a copy of the questionnaire was sent to each subject who had been allotted to Form 1 and not returned the questionnaire.Those who did not respond to the second questionnaire within 20 days after sending the reminder were called and given interviews.This telephone strategy entailed up to 10 calls in a twoweek period.
Since the intention of the trial was to determine what kinds of questionnaires are available for the populationbased mail survey, the primary outcome was cumulative response rates.We compared the numbers of responses before the second mailing of Form 1 among the six types of questionnaires.
We classified the subjects of Form 1 into three groups of early respondents (within 18 days of the initial mailing), late respondents (after 18 days), and nonrespondents (respondents to telephone strategy), and compared the characteristics of them in relation to cigarette smoking and drinking and environmental exposure to solvents.Respondents' characteristics associated with the type of questionnaire were also assessed for the variables commonly contained in the forms.

RESULTS
Table 2 shows that response rates to Form 1, 2, 3, and 4 were similar and somewhat higher than those to Form 5 and 6, although they did not attain statistical significance.It might be due to relatively small numbers of samples.Approximately 90% of the respondents returned the questionnaire within 10 days.
For Form 1, 54 persons returned the first mailing questionnaire, 19 returned the second mailing questionnaire.Five persons were notified to be dead.We could reach all of 22 nonrespondents (or family members) except one by telephone, of whom 16 cooperated for the telephone interview, 1 refused, I was disabled, and 3 were notified to have moved alone on business.After excluding ineligibles (dead, disabled, and staying out for long term on business) from the denominator, we calculated a response rate of 80.2% (=73/91) after two mailings.
Table 3 indicates that cigarette smokers and drinkers are less represented in the respondents group and ex-smokers respond quickly.The proportions of current cigarette smokers and drinkers to nonrespondents were significantly higher than those to respondents (p<0.01,respectively).Although late respondents were more likely to be current cigarette smokers or drinkers than early respondents, the differences were not substantial.Numbers of cigarettes smoked per day and years of smoking, and age at cigarette smoking started were not different among the current cigarette smokers in the three groups.The proportion of heavy drinkers with more than 100 ml/day of ethanol consumption to current drinkers was significantly higher in nonrespondents than in early respondents.Environmental exposure to solvents were more prevalent in later respondents or nonrespondents, but the number is too small to test the statistical significance.Elder subjects tended to respond more quickly but age distributions of three groups did not vary substantially.The number of the questions answered were similar among early, late respondents and nonrespondents.

DISCUSSION
The study subjects were ristricted to the male oweners of telephone listed in the telephone directory.Although the prevalence of telephone ownership in Gifu City is thought to be high (telephone coverage rate is not informed in the public), our procedure of sampling does not approximate a true random sampling of the population.The owners listed in the telephone directory are likely to be old and relatively high proportion of deceased persons among the subjects might be due to this sampling frame.However, our findings provide basic imformation for conducting similar mail surveys.
We achieved a good response rate of nearly 70% on the second mailing for Form 1.
As questionnaire length extends as we add items, we cannot completely control the effect of either of two variables, i.e., length and topics, when the difference in response rates is observed.However, additional questions on past medical history in Form 2 or family medical history in Form 3 did not reduce the response rate which was obtained for Form 1 asking basic life style practices and environmental exposure to solvents.Without need to control the effect of length, we noted that these findings did not support that the topics of a mail questionnaire can have a profound influence on response rates.The question on consanguineous marriage were considered to be so sensitive that the subjects might be reluctant to respond whether they have such a history or not.Although this topic did not reduce the entire response rate, the sample size was too small to measure the influence of this question on response from the subjects with a history of consanguinous marriage which are thought to be relatively small group (1.3% of the respondents reported the history of consanguinous marriage between parents or grandparents).
Few studies examined the effect of the topics of questionnaire.Kaplan et al.10) found that addition of a question on schooling decreased the response rate from the elder women living in a city.Review article by Heberlein et al.11> presented that questions about social economical status, ownership, work organization, or household did not lower or raise response rates.
Previous studies concluded that questionnaire length would not affect the response rate, but they included relatively longer questionnaires in their design6,7).When both past medical and family histories were added to the shortest questionnaire in Forms 5 and 6, somewhat lower response rates were noted in our study.Although the reduction was not substantial, we cannot deny the possibility that much longer questionnaire with more items might decrease the response rate.
A serious question remains as to the representativeness of acquired sample when high response rate is not achieved in mail survey.As the mail strategy and the telephone strategy with intensive follow-up of nonrespondents covered 89 percent of the samples to Form 1, we were able to compare the characteristics of respondents and nonrespondents to mail questionnaire.Cigarette smokers and drinkers were under-represented in the respondents.In addition, cigarette smokers and drinkers were likely to respond more slowly, while ex-smokers were likely to respond more quickly.Similar results concerning cigarette smoking were noted in several previous studies [12][13][14][15] In terms of drinking status, Cottler et al. 16) showed that the estimate prevalence rate of alcoholism increased with the number of contacts attempts.
Although we did not separate the effects of cigarette smoking and alcohol drinking on the response because most of cigarette smokers were drinkers, these findings give us the suggestion to characterize the nonrespondents.They may hesitate to report their unhealthy life habits such as smoking and drinking.They may be not interested in research topics of health survey, as their smoking or drinking status may reflect the lower level of concern to the health.
We found some indication that younger subjects were slower to respond.Some studies found decreasing response with increasing age10.15)and others found little relation between age and response.16,17) As the attempts to reach nonrespondents were limited to the samples for Form 1, the characteristics of them which we can compare to those of respondents did not cover other health status.We compared the early and late respondents on the variables for past medical and family histories and found no substantial differences between them.
We addressed two main aspects to examine the usefulness of mail questionnaire ; the adequacy of response rate and bias resultant from non-response.As approximately 70% of response rate was not enough to avoide the nonresponse bias with respect to smoking and drinking, it was concluded that one repeated mailing of questionnaires could not be useful for surveys dealing with self-selected status such as smoking and drinking.This study also examined whether the two factors of questionnaire design, topic and length, would affect the response rate.Although we may have lacked statistical power to detect some differences in response rates in our study, the effects of these factors on the response rate were not substantial.This result may not be generalized because we did not include very sensitive topics and the lengths we compared were at most three pages.However, the topics we chose for the mail questionnaire were commonly searched in the epidemiological studies, although some of the questions were specific to hematopoietic disorders.
Our study may show the standard response rate from the telephone-based samples in a city with middle size of population in Japan and be useful to other investigators concerned about the effectiveness of mail survey.

Table 1 .
Topics and length of the six forms of questionnaires.

Table 2 .
Cumulative number of mails returned by type of miestinnnaires

Table 3 .
Age, cigarette smoking, alcohol drinking, and environmental exposure to solvents by respondent status.