THE EFFECT OF DATA COLLECTION MODE AND ETHNICITY OF INTERVIEWER ON RESPONSE RATES AND SELF-REPORTED ALCOHOL USE AMONG TURKS AND MOROCCANS IN THE NETHERLANDS: AN EXPERIMENTAL STUDY

Abstract

Aims: To test the effects of data collection mode and ethnicity of interviewers on response rates and self-reported alcohol use among second-generation Turks and Moroccans in Rotterdam, The Netherlands. Methods: Two hundred and sixty-nine Turks and 271 Moroccans were interviewed face-to-face, and 475 Turks and 482 Moroccans received a mailed questionnaire. Half of the Turks and Moroccans randomly allocated to the interview mode were ethnically matched to the interviewer; the remainder were allocated to a Dutch interviewer. Results: Turks and Moroccans more often responded to a face-to-face interview than to a mailed questionnaire. No effect of ethnicity of interviewer on response rates was demonstrated. With respect to the effects on alcohol reports, Turks and Moroccans tended to report a higher alcohol use in the mailed survey than in the face-to-face interview. They reported significantly more often excessive drinking in the mail survey than in the face-to-face interviews. Ethnicity of the interviewer resulted in Turks and Moroccans reporting a higher prevalence of alcohol use during the previous 6 months when interviewed by a Dutch interviewer compared with an ethnically matched interviewer. Conclusions: Among second-generation Turks and Moroccans, mail surveys seem most suitable to measure mean and excessive alcohol use. However, interviews held by Dutch interviewers seem to be the most appropriate method to study the prevalence of alcohol use during the previous 6 months.

( Received 21 January 2004; first review notified 22 March 2004; in revised form and accepted 29 July 2004; Advance Access publication 29 March 2005 )

INTRODUCTION

Turks and Moroccans form two of the largest non-western migrant groups in the Netherlands and constitute ∼13% of the Rotterdam population, the second largest city in the Netherlands ( Bik and Stolk, 2002 ; CBS, 2002 ). Studies on alcohol prevalence in the Netherlands reveal that the proportion of drinkers is lower among Turks and Moroccans, compared with the autochthonous Dutch population ( Abraham et al ., 2000 ; Planije et al ., 2000 ; van Eijsden et al ., 2004 ). However, on account of the Islamic background of Turkish and Moroccan immigrants, which prescribes alcohol abstinence, it is assumed that alcohol use is substantially underreported by these groups ( Gorissen et al ., 1988 ). Therefore, the reliability of these alcohol figures has been questioned.

Alcohol research among migrant groups with an Islamic religious background faces several methodological problems, which may generate unreliable alcohol data for these groups. First of all, probably owing to their Islamic religious background, Turks and Moroccans are often underrepresented in alcohol studies ( Planije et al ., 2000 ). When a sampled person does not respond to a survey request, non-response occurs, which may lead to biased results ( Dillman et al ., 2002 ). Therefore, reducing non-response is a prerequisite for reliable data collection.

In general, non-response is higher in mail surveys compared with face-to-face and telephone interviews ( Hox and de Leeuw, 1994 ). This effect may result from the fact that interviewers can persuade people to participate in the study. However, to some extent, this effect also seems to depend upon the subject of study. For instance, if the subject of study concerns a socially undesirable behavior, because of the higher perceived anonymity people may be more willing to participate in mail surveys than in face-to-face interviews ( Schwartz et al ., 1991 ; Gmel, 2000 ). This may apply in the Netherlands where Turks and Moroccans may view alcohol use as proscribed by their community.

In the case of face-to-face interviews, the ethnic background of the interviewer may also influence response rates ( Weeks and Moore, 1981 ). Compared with an interviewer of a different ethnic background, an interviewer with the same ethnic background as the respondent may attain higher response rates, simply because the participants are more at ease with the interviewer. Interviewers with the same ethnic background would more easily elicit feelings of trust ( Dotinga et al ., 2004 ), which may increase the willingness to participate in a study ( Dillman et al ., 2002 ). However, empirical data on response rates in alcohol studies do not always support the idea that ethnic matching would enhance response rates. For example, ethnic matching of interviewer and respondent did not increase response rates in a survey on legal and illegal drug use among Moroccans in Amsterdam ( Abraham et al ., 1999 ). In addition, two studies in the Netherlands showed that a relatively large number of Moroccans were willing to participate in a study in which a Dutch interviewer questioned them about alcohol use ( Kemper, 1998 ; Planije et al ., 2000 ). Thus, it remains unclear whether face-to-face interviews or mail surveys will elicit higher response rates in a study on alcohol use among Turks and Moroccans in the Netherlands. Moreover, the literature is not clear as to whether ethnically matched interviewers generate higher response rates. Therefore, the present study examines the effect of data collection mode and ethnicity of the interviewer on response rates in alcohol research.

In addition to the effect on response rates, the data collection mode and ethnicity of interviewers may also have an effect on alcohol reports. However, studies on the effect of data collection mode on self-reported alcohol use have yielded contradictory results. For example, Bongers (1998) showed that alcohol reports did not differ between mail surveys and personal interviews among inhabitants of Rotterdam. Aquilino (1994) found that an admission of alcohol use was most likely in self-administered questionnaires and less likely in face-to-face interviews. However, a study among young Moroccans in the city of Utrecht (The Netherlands) showed that alcohol reports were higher in face-to-face interviews than in self-administered questionnaires ( Planije et al ., 2000 ). Based on these studies, it remains unclear whether different data collection modes will produce differences in alcohol reports and if so, which data collection mode will yield more reliable alcohol data. In the present study, reports of higher alcohol consumption are considered to be the more reliable and accurate data. This assumption is generally made in alcohol research among the total population, because estimates of total alcohol consumption based on self-reports often cover only 40–60% of the alcohol consumption based on other indices of alcohol use, such as sales data ( Lemmens et al ., 1988 ; Rehm and Spuhler, 1993 ). Because the religious and cultural background of Turks and Moroccans prescribes the abstinence of alcohol use, Turks and Moroccans can be expected to underreport their alcohol use to an even larger extent than the autochthonous population. Therefore, the assumption that higher alcohol reports constitute more accurate data is particularly tenable in research among Turks and Moroccans.

With respect to face-to-face interviews, the ethnicity of the interviewer may also influence alcohol reports. This is evident when respondents are questioned about culturally sensitive issues ( Weeks and Moore, 1981 ) and when respondents believe that certain answers are not socially acceptable ( Van 't Land, 2000 ). Extensive research conducted in the USA has addressed ethnicity-of-interviewer effects and validity of self-reports (e.g. Campbell, 1981 ; Weeks and Moore, 1981 ; Anderson et al ., 1988 ; Davis, 1997 ), indicating that these effects may be explained by the salience of the ethnicity of the interviewer throughout the interview and the tendency of the respondent to answer according to the racial attitudes of the interviewer. If these are indeed the operating mechanisms, Turks and Moroccans can be expected to be more willing to report alcohol use to a Dutch interviewer, since Dutch people generally approve of ‘responsible’ alcohol use whereas Turkish and Moroccan people generally disapprove of any form of alcohol use. Furthermore, social control mechanisms in Turkish ( Ögel, 1997 ) and Moroccan societies ( Van Gemert, 1998 ) may explain higher alcohol reports to Dutch interviewers. These social control mechanisms tend to discourage disrespectful behavior with respect to family and community norms, e.g. to abstain from alcohol. Therefore, Turks and Moroccans may be more willing to respond to alcohol questions asked by a Dutch interviewer than by an ethnically matched interviewer.

The present study investigates the above mentioned methodological issues using an experimental design. More specifically, the following research questions will be addressed with respect to alcohol research among Turks and Moroccans in the Netherlands: (i) which method will yield higher response rates, mail questionnaires or face-to-face interviews? (ii) which interviewer will attain higher response rates, a Dutch or a Turkish/Moroccan interviewer? (iii) which method will yield higher alcohol reports, mail surveys or face-to-face interviews? (iv) which interviewer will attain higher alcohol reports, a Dutch or a Turkish/Moroccan interviewer?

MATERIALS AND METHODS

Sample

To decide upon the sample size needed for a large power of the study, the data of an alcohol prevalence study among Moroccans in the age of 16–35 years, living in Utrecht was used ( Planije et al ., 2000 ). Based on these data, the present study would have elicited responses of 114 Moroccans in the interview mode (response rate of 42%) of which 30 Moroccans would have been drinkers (26%) and it would have elicited responses of 149 Moroccans in the mail survey (response rate 31%) of which 24 Moroccans would have been drinkers (16%).

From the municipal register of Rotterdam, 744 second-generation Turks and 753 second-generation Moroccans (≥16 years) were randomly selected from four neighborhoods (selected for their geographic distribution) in Rotterdam. To gain insight into the impact of the data collection mode on response rates and alcohol reports, a random sample of 269 Turks and 271 Moroccans was selected to be interviewed face-to-face. Half of the Turks and Moroccans allocated to the face-to-face condition were ethnically matched to the interviewer and the other half was matched to a Dutch interviewer. The remainder of both the samples (475 Turks and 482 Moroccans) received a mailed questionnaire. All sample selections were stratified for gender and neighborhood.

From the original sample of 744 Turks, 14 respondents were excluded from analyses because they belonged to the first-generation Turks, one Turk had completed the same questionnaire twice, and another one had completed the questionnaire on paper at the door, whilst he should have answered the questions face-to-face. The final sample consisted of 728 Turks (463 in the written data collection mode, and 265 in the interview mode): of these, 385 were men and 343 were women, with an age range of 16–34 years ( M = 22). From the total sample of 753 Moroccans, eight respondents were excluded from analyses because they were born in Morocco (first-generation Moroccans). The final sample consisted of 745 Moroccans (481 in the written data collection mode, and 264 in the interview mode): of these, 360 were men and 385 were women, with an age range of 16–34 years ( M = 20.8).

Procedures

Interviewers were recruited among Turkish and Moroccan university students belonging to the second generation ( Dotinga et al ., 2004 ). Ten Dutch interviewers (of which four were male, mean age of 24.5 years, and six were female, mean age of 23.6 years), eight Turkish interviewers (of which three were male, mean age of 23.3 years, and five were female, mean age of 23.6 years) and six Moroccan interviewers (of which three were male, mean age of 20.3 years and three were female, mean age of 20.3 years) participated in the study. Interviewers and respondents were gender matched. Only female students who did not wear a headscarf were selected as interviewers. Wearing a headscarf is a symbol of the Islamic religion and since abstention from alcohol use is an Islamic prescription rule, this might have influenced the answers about alcohol use of respondents. Since both the interviewers and the interviewees comprised second-generation Turks and Moroccans (i.e. Turks and Moroccans born in the Netherlands) interviews were performed in Dutch. To prevent interviewers from interviewing people with whom they were acquainted, interviewers were not allowed to work in the neighborhood where they lived themselves. Interviewers attended an extensive training session, in which information was given on the background of the study, the protocol that had to be followed and techniques that had to be used. For example, interviewers were instructed to interview respondents without the presence of others, e.g. in a separate room or somewhere else. If this was not possible they had to make another appointment at a time in which the respondent could be interviewed alone. Furthermore, the questionnaire was discussed and the interview was practised by means of role playing.

Turks and Moroccans selected for face-to-face interviews received a letter which explained the study and also announced that an interviewer would visit them in the coming weeks. Interviewers were instructed to approach an interviewee at home at least three times at different time points. Turks and Moroccans who were allocated to the written data collection mode received a mailed questionnaire at the same time that the interviewers started the interviews. To increase the response rate and to keep the procedures for both data collection modes as similar as possible, non-respondents to the mail survey received two reminders after 3 and 6 weeks, respectively. Both reminders included another copy of the same questionnaire. Four months after the start of the data collection process, non-respondents to the mail survey and the face-to-face interviews received a shortened mailed questionnaire.

Measurements

Alcohol use was measured according to the quantity– frequency–variability method (QFVmethod) using six questions: (i) ‘Which alcoholic drinks did you use during the past six months’ (beer, wine, strong alcoholic beverages, I haven't drunk alcohol during the past 6 months, I have never drunk alcohol); (ii) ‘How many weekend days (Friday through Sunday) do you drink on average’ (3 days, 2 days, 1 day, occasionally, I never drink on weekend days) (F); (iii) ‘How many glasses do you drink on average on a weekend day’ (>11 glasses, 7–10 glasses, 6 glasses, 4–5 glasses, 3 glasses, 2 glasses, 1 glass) (Q); (iv) ‘How many weekdays (Monday through Thursday) do you drink on average’ (4 days, 3 days, 2 days, 1 day, occasionally, I never drink on weekdays) (F); (v) ‘How many glasses do you drink on average on a weekday’ (Q); (vi) ‘Have you ever drunk six or more glasses in 1 day during the past 6 months?’ (every day, 5 or 6 times a week, 3 or 4 times a week, 1 or 2 times a week, 1–3 times a month, 3–5 times per six months, 1 or 2 times per 6 months, never) (V).

Respondents were classified as abstainers if they reported not having had a drink in the previous 6 months. Weekly alcohol use based on the QF items was assessed by the sum of alcohol use on weekdays (number of drinking weekdays × number of glasses on a weekday) and weekend days (number of drinking weekend days × number of glasses on a weekend day), (test–retest reliability of this QF scale was r = 0.77; P < 0.01). If respondents reported (both on weekdays and weekend days) only occasional drinking they were classified as occasional drinkers, otherwise they were classified as regular drinkers. Weekly alcohol use based on the variability item was assessed by multiplying the number of days per week that a person drank at least six glasses of alcohol × six glasses. When an answering category included a range, the mean of that range was taken and multiplied by six (e.g. 1–3 times a month: mean is 2 times a month = 0.5 times a week, multiplied by six = 3). When weekly alcohol use according to the QF items was inconsistent with weekly alcohol use according to the V item, the highest score was taken as the indicator for weekly alcohol use. Respondents were defined as excessive drinkers according to two definitions: (i) based on the variability item: if they reported drinking at least once a week six or more glasses of alcohol ( Garretsen, 1983 ), or (ii) based on the QF items: if women reported drinking >13 glasses of alcohol a week and men reported drinking >20 glasses of alcohol a week (e.g. San José, 2000 ; Toet et al ., 2003 ).

Socio-demographic factors measured in this study were gender, age, marital status (having a partner or not) and education. Education was defined as the respondent's highest attained education level: (i) primary school, (ii) lower vocational/general, (iii) intermediate vocational/general and (iv) higher general/higher vocational/university.

Analyses

First, the distribution of background factors was compared between the respondents to the different data collection modes. Statistical significance for these analyses was estimated by Chi-square test. Second, the effect of data collection mode (total sample) and ethnic background of the interviewer (interviewed sample) on response rates was tested. To determine statistical differences, logistic regression models were fitted, including age and gender. Third, the effects of data collection mode (total sample) and ethnic background of the interviewer (interviewed sample) were tested on alcohol reports. Logistic regression models were fitted (including age, gender, marital status and educational level) to test the effect of data collection mode and ethnicity of interviewer on the number of drinkers and on excessive drinking. For the latter, a distinction was made between regular and occasional drinkers, since occasional drinkers could distort the results with respect to the quantity of alcohol use. Variance analyses (including age, gender, marital status and educational level) were conducted to test the effect of data collection mode and ethnicity of interviewer on mean alcohol use in both the total sample of drinkers and among the regular drinking sample.

RESULTS

Demographics

Table 1 presents differences in background factors between the respondents to the mail survey and the respondents to the interview. Significant differences were found for gender and education. Relatively fewer men responded to the mail survey than to the interview (χ ² = 5.52; df = 1; P < 0.05). Furthermore, respondents with the highest educational level more often responded to the mail survey than to the interview (χ ² = 16.52; df = 3; P < 0.01). To control for differences in age, gender, marital status and educational level, statistical analyses testing the effect of data collection mode and ethnicity of interviewer on response rates and alcohol reports were corrected for these four socio-demographic factors by including these factors as covariates in the analysis.

Response rates

The overall response rate for the mail survey was 37.6% ( n = 355) and for the face-to-face interview 42.9% ( n = 227) ( Table 2 ). The odds of participating in the face-to-face interviews were significantly higher than the odds of participating in the mail survey [OR = 1.26; 95%CI (1.01–1.57)]. Table 2 also presents the associations between ethnic background of the interviewer and response rates. No significant differences in response rates were found between Dutch and Turkish/ Moroccan interviewers.

Respondents and non-respondents to the interview differed significantly in age: fewer respondents aged 22–27 years responded to the interview than in the younger (16–21 years) and older (28–34 years) age groups (χ ² = 9.71; df = 2; P < 0.01). However, no significant differences in self-reported alcohol use between respondents and non-respondents within each age category were found. Moreover, no significant differences in alcohol use between respondents and non-respondents to the interview were found. Respondents of the mail survey differed from the non-respondents on gender: significantly more women responded to the mail survey compared with men (χ ² = 33.29; df = 1; P < 0.001). However, no significant differences in self-reported alcohol use between respondents and non-respondents were found within each gender category. Moreover, among both Turks and Moroccans, no significant differences in alcohol use were found between respondents and non-respondents of the mail survey.

Self-reported alcohol use

With respect to the effect of data collection mode on the number of drinkers ( Table 3 ), no significant differences were found between respondents of the mail survey and respondents of the interview. With respect to the effect of ethnicity of interviewers on the number of drinkers, data showed that respondents who were interviewed by a Dutch interviewer had significant higher odds of reporting alcohol use in the previous 6 months, compared with respondents who were interviewed by a Turkish/Moroccan interviewer [OR = 3.05, 95%CI (1.51–6.18)]. Correcting for the presence of others by including this factor as a covariate in the regression model did not change these results.

Table 4 presents the effects of data collection mode and ethnicity of the interviewer on the average number of drinks per drinker. For both the total drinking sample and the regular drinking sample no significant differences were found in mean alcohol use between data collection modes. However, in both samples, data point in the same directions, i.e. alcohol consumption was higher in the mail survey compared with the interview mode. Furthermore, for both the total drinking sample and the sample of regular drinkers, no significant differences were found in average alcohol use between Dutch and ethnic matched interviewers. Again, although not significant, the mean scores point in both samples in the same direction, with lower means for alcohol use among respondents to the Dutch interviewers.

With respect to excessive drinking among regular drinkers ( Table 5 ), significant differences between data collection modes were found when excessive drinking was defined as >13 glasses of alcohol in 1 week for women and >20 glasses of alcohol in 1 week for men [OR = 0.61, 95%CI (0.00–1.00)]. Respondents of the interview significantly less often reported excessive drinking compared with respondents of the mail survey. Finally, because of the small numbers, the effect of ethnicity of the interviewers on excessive drinking could not be tested.

DISCUSSION

Our principle findings were as follows: A few limitations with respect to the external validity of our findings should be addressed. First, some differences in age and gender between our respondents and non-respondents were found. These differences would decrease the external validity of the present study if they are related to alcohol use. However, additional analyses generated no differences in self-reported alcohol use between respondents and non-respondents within each of the gender and age categories. Therefore, differences in the socioeconomic characteristics of respondents and non-respondents are not expected to have seriously biased our results. Some caution is needed in generalizing the results of the present study to the general population of Turks and Moroccans in the Netherlands.

1. Mode of data collection affected response rates: Turks and Moroccans responded more often to a face-to-face interview compared with a mail survey. This result is in accordance with a meta-analysis which analyzed response rates obtained by different data collection modes in 45 studies ( Hox and De Leeuw, 1994 ), and can be explained by the fact that interviewers can persuade people to participate in a study. It is often assumed that this could hold particularly true for abstainers, among which non-response rates are often relatively high ( Dillman and Carley Baxter, 2000 ; Lahaut et al ., 2002 ). Interpersonal contact would give the interviewer the opportunity to explain to abstainers that their participation is as relevant as the participation of drinkers. However, the present study does not support this notion because, the numbers of abstainers in the mail survey and the face-to-face interview condition do not differ.

2. Ethnicity of the interviewer did not affect response rates.

3. Mode of data collection did not affect the number who reported they were drinkers. However, excessive drinking was reported in the mail survey more often than in face- to-face interviews. Although not significant, data with respect to mean alcohol use point in the same direction with a higher mean alcohol consumption admitted in the mail survey than in the face-to-face interviews. The findings with respect to excessive drinking and average reported alcohol use may result from the fact that anonymity is valued more in mail surveys than in face-to-face interviews ( Schwartz et al ., 1991 ).

4. Ethnicity of interviewers affected the number who reported they were drinkers, more drinkers reporting that they had used alcohol in the past 6 months when interviewed by a Dutch interviewer, compared with an ethnically matched interviewer. Turks and Moroccans may more easily report alcohol use to a Dutch interviewer and may exaggerate their alcohol use to comply with the norms of the Dutch interviewer ( Van 't Land, 2000 ). This may be because reporting alcohol use is perceived as being more socially acceptable to a Dutch interviewer than to an ethnically matched interviewer. On the other hand it may also be explained by religious motives. Islamic people might deny their alcohol use more towards interviewers with the same religion, a religion that prescribes alcohol abstinence. A third mechanism which may underlie this finding is that Dutch interviewers may have had better interviewing skills than Turkish and Moroccan interviewers, and may have questioned the participant's statement of not having used alcohol in the past 6 months, thereby provoking alcohol reports of very occasional drinkers. Indeed, the results of the present study showed that the number of occasional drinkers is particularly high in the Dutch interviewer condition. However, the results with respect to differences in mean alcohol use between interviewers point in the opposite direction, showing that alcohol reports were higher among respondents interviewed by ethnically matched interviewers, even when occasional drinkers were left out from analysis. Thus, when respondents decided to report alcohol use to an interviewer from their own ethnic group at all, no resistance was felt in reporting high quantities of alcohol use.

Third, we want to comment on the response rates in our study. Although response rates are higher in face-to-face interviews compared with mail surveys, response rates are low in both data collection modes. However, it is comparable with other Dutch alcohol studies (e.g. Bongers and van Oers, 1998 ; Planije et al ., 2000 ; Lahaut et al ., 2002 ) and is partly explained by the fact that participation in governmental surveys is not obligatory ( De Heer, 1999 ). Furthermore, people may be tired of participating in surveys, because they are approached very often. Fourth, conducting interviews in participants' homes may have had some adverse impact on attaining 'private' accounts ( Malseed, 1990 ) of alcohol consumption. However, this was necessary to keep the experimental conditions of both data collection modes the same. We tried to rule out the effect by instructing the interviewers that they had to interview people alone, without the presence of others.

Finally, we want to comment on the way in which the data were analyzed. It has been argued that Turks and Moroccans, although both Islamic oriented, are different in their cultural and religious background. Therefore, studying these two ethnic groups simultaneously would generate oversimplified insights. However, since power problems were evident when analyzing Turks and Moroccans separately, it was decided to combine both samples in the analysis. Moreover, analysis performed separately among Turks and Moroccans, revealed effects in the same directions for both ethnic groups, supporting our analysis among the combined sample.

Based on the results of the present study, we have to draw different conclusions about the most appropriate way to measure alcohol use among second generation Turks and Moroccans in the Netherlands, depending on the goal of the study. To measure the prevalence of alcohol use in the previous 6 months, face-to-face interviews with Dutch interviewers seem most appropriate, but to measure mean alcohol consumption levels and excessive drinking, mail surveys seem more appropriate.

This research was funded by the Netherlands Health Research and Development Council.

REFERENCES

Author notes

1Addiction Research Institute Rotterdam (IVO), Rotterdam, 2The Amsterdam Bureau for Research and Statistics (O+S), Amsterdam and 3Tilburg University, Faculty of Social and Behavioural Sciences, Tilburg, The Netherlands