Response Bias by Neuroblastoma Screening Participation Status and Social Desirability Bias in an Anonymous Postal Survey , Ishikawa , Japan

Objective: To examine response bias by neuroblastoma screening participation status in a population-based postal survey of parents in Ishikawa Prefecture, Japan Methods: The eligibility criteria for the study were: 1) parents whose infants were born in Ishikawa Prefecture between March 1997 and February 1998, and 2) of those parents who resided in the Prefecture in March 1999. Four-page questionnaires were mailed to one-third of screening participants (n= 2,886) and all the nonparticipants (n= 1,401). Questionnaires were anonymous, with no identifiers on the questionnaire. Colored papers were used for printing questionnaires to differentiate screening participation status. Response rates were calculated using demographic information on the infant registry as the denominator and demographic characteristics data from the returned questionnaire as the numerator. Results: The response rate was 63% for participants and 33% for nonparticipants. The following factors were associated with lower response rates regardless of screening participation status: older maternal age (*35 years), higher parity (*4), nuclear family status, and mother having a full-time occupation. Approximately 20% of screening nonparticipants reported having participated in the screening. Place of residence, maternal age, and parity were associated with the percentage of incorrect reporting. Conclusion: Screening participation status was a major factor associated with low response rate, although some demographic characteristics were also predictive of low response rates. Incorrect reporting of screening participation among nonparticipants indicates a strong social desirability bias in this official survey in Japan. J Epidemiol, 2001 ; 11 : 70-73

*Overall rates were adjusted for sampling .
Response Bias and Social Desirability Bias Notes: Because of missing data for denominators, ratios may exceed 100%.
among participants.Maternal age * 35 was a risk factor for low response ratios among nonparticipants (Table 2).

Maternal occupation
Full-time employment was predictive of lower response ratios, especially for screening participants (Table 2).

Family structure
Nuclear family status was associated with lower response ratios than extended family status regardless of participation status.
Except for parity, a few percentages of demographic data were missing on the registry, and 60% to 80% of the missing demographic data belong to Kaga City (N= 408).After excluding the Kaga City data, response ratios changed by a few percentage points in most of the strata.

Self-reported screening participation status compared to the screening registry
Five out of 1,826 screening participants reported not having participated in the screening.Of these five, four mothers were multiparous.In contrast, 21% of nonparticipants replied they had participated in the screening.The incorrect reporting rates varied greatly among the five districts (Table 3).Other factors associated with the proportion of incorrect reporting were lower parity and maternal age (Table 3).Further stratifications were not possible due to the small cell size.

DISCUSSION
This study demonstrated the usefulness of colour-coded questionnaires to detect response bias in an anonymous survey.This method enabled us to calculate accurate response rates by screening participation status.If the responses on the questionnaire were taken at face value, the response rates would have been 25% for nonparticipants and 66% for participants.
This method also enabled us to calculate the proportion of mothers providing incorrect information regarding screening participation status.It is not clear why some nonparticipants reported to have participated in the screening in the anonymous survey.Since no tracing was used, mothers did not have to respond if they did not wish to participate in the survey.It is unlikely that mothers were confused about screening participation status with their older children's screening participation because primiparous mothers had the highest percentage of incorrect screening participation status.This problem of incorrect reporting may reflect a strong social desirability bias in a conservative society.Mothers may feel psychological pressure to return the questionnaire in an official survey and may also feel uncomfortable reporting nonparticipation status.In this survey, no questionnaire was returned unanswered, which is a form of refusal reported in the literature 5).
Response ratios exceeded 100 in some strata, possibly due to missing data for denominators, changes in status, or problems in recalling demographic information.We could not evaluate the accuracy of responses.Nevertheless, patterns of response ratios by demographic characteristics are in the expected direction.Higher parity, older maternal age, and full-time employment status were associated with lower response ratios.These demographic characteristics may reflect the degree of maternal interest in the screening or the priority of the postal survey in their lives.
Previous studies have indicated that educational attainment is associated with response rates 6,7).In Ishikawa Prefecture, over 99% of the population receives a high school education 8).
This survey did not request information regarding maternal education; it was thought that asking about educational status would lower response rates in this population.Regional variations in the percentage of students who receive a college education were not correlated with response rates by region 8).
Kanazawa has the highest percentage of people with a college education, and the response rate for Kanazawa was lower than that for Noto-North, where the percentage of people with advanced degree is much lower than the prefecture average, having the highest response rate 8).It seems that local culture is a stronger predictor of response rates than educational attainment in Ishikawa Prefecture.
Asch and colleagues conducted a systematic review of postal surveys reported in the medical literature and found that anonymity and financial incentives were not associated with higher response rates 2).In this study, it was a consensus of the planning committee that maintaining anonymity was important to ensure cooperation in this population.Multivariate analysis of published literature results suggests written reminders with a copy of the questionnaire and telephone reminders were each associated with a response rate about 13% higher than surveys that do not use these tools 2).This study could not test the effect of reminders to increase response rates because of budget constraint.In any event, a 10% increase in response rate for screening nonparticipants would not have met the goal of the survey, i.e., to explore reasons for nonparticipation.In summary, colour-coded questionnaires detected severe response bias and social desirability bias in a population-based official survey in Japan.Screening nonparticipants were less likely to respond than screening participants, and 20% of nonparticipants identified themselves as screening participants.

INTRODUCTIONA
neuroblastoma screening program was instituted in Japan in 1985.The screening is offered free of charge by the Prefectural government when parents bring their children for health and developmental check-ups at 3 months of age.Municipal offices provide a container, an addressed envelope, and a leaflet explaining how to collect a urine sample.Parents are requested to send the specimen to the laboratory when the child reaches 6 to 8 months of age.Each municipality sends the Public Health Department (PHD) a list of infants and their addresses.The five PHDs maintain a neuroblastoma screening registry, which includes mother's demographic information, screening participation status, and lab results.

Table 1 .
Response rates by public health district and screening participation status.
The main laboratory sends the results to the PHDs, and the PHDs notify parents if further tests are necessary.providedanopportunityto study response bias in a populationbased survey of parents by infant screening participation status.and about one-third of the participants were chosen using systematic sampling to select every third participant on the list within each PHD.The four-page questionnaire included 22 items.Twelve items solicited demographic information and screening participation status; 6 items for screening participants asked reasons for participation and problems associated with screening procedures; 4 items for nonparticipants asked reasons for nonparticipation.The questionnaire also requested the following demographic information: family structure, birth order of the infant, maternal age at the time of the survey, and maternal occupation when the infant was 6 to 8 months old.The questionnaire with a cover letter was mailed in March 1999, and those returned within three weeks were included in the analysis.This was an anonymous postal survey, and colored papers