Comparative Appraisal of Randomized Response Technique vs. Direct 1 Interview Method in Assessment of Burden of Tobacco and Alcohol Use 2 among Adolescents in India

27 Background: In countries like India, the major source of error in reporting of sensitive events 28 (e.g., tobacco and alcohol use) among adolescents is deliberate misreporting. To estimate its 29 burden, the commonly used direct interview method involves problems in reporting. To cope up, 30 randomized response technique (RRT) is an alternative approach which uses a random device for 31 providing more privacy to respondents. So, it may be worthwhile to compare RRT with direct 32 interview method in assessment of burden of ever tobacco and alcohol use among Indian 33 adolescents. Methods: A cross-sectional survey on ever tobacco and alcohol use among adolescents was 35 conducted on 796 (401: RRT; 395: Direct interview) 9 th to 12 th class students from purposively 36 selected three schools of Delhi/National Capital Region, during November-2014 to November- 37 2016. As per conventional use of RRT, two types of questions, one related to “sensitive 38 attribute” and other “an unrelated question” was used to estimate ever tobacco and alcohol use. 39 For the first time, a new random device (i.e., randomly arranged questionnaires) was proposed 40 and used under RRT which is user friendly and appropriate for time-bound application. For 41 comparison, the estimates of both, ever tobacco use and ever alcohol use, were obtained which 42 could also facilitate comparative change in effect size with increasing sensitivity. 43 Results: The prevalence of ever tobacco use under RRT approach and Direct Interview method 44 was 18.6% (95% CI: 13.33-24.01) and 10.1% (95% CI:7.15-13.10) respectively, where as that of 45 ever alcohol use was 22.8% (95%CI: 17.08-28.5) and 9.1% (95% CI: 6.27-11.95) respectively. Further, comparative results showed that as contents in used substance become more sensitive, 47 under estimation of burden under direct interview method may become higher [i.e., Ever tobacco 48 use: 8.5% (95% CI: 3.43-13.65); Ever alcohol use: 13.6% (95% CI: 8.34-19.03)]. The proposed 49 new random device under RRT, as a set of randomly arranged questionnaires containing either 50 sensitive or unrelated questions, consists of most of its recommended properties. 51 Conclusions: The findings under the present study suggest that the randomized response 52 technique may serve as a versatile method for gaining access to more accurate information on 53 sensitive topics.

There is ample evidence in the related literature that tobacco and harmful alcohol use are major 60 leading threats for various non-communicable diseases and death globally. There are numerous 61 diseases like different types of cancer, respiratory diseases, and heart diseases, which are known 62 to be caused and get further deteriorated by use of such substances. As a matter of fact, these bad 63 practices are major modifiable causes of morbidity and mortality. Further, their uses among 64 adolescents are likely to have many grisly impacts on their future life, and society. Also, tobacco 65 and alcohol use are called as gateway regarding initiation of illicit drug like amphetamine, 66 cocaine, hallucinogens or heroin [1]. Hence, the time to time assessment of the burden of 67 substance use, like tobacco and alcohol use, among adolescents becomes more important. 68 However, such tasks among adolescents are often tedious and some time controversial. The 69 4 surveys using direct interview methods on sensitive personal practices (e.g., tobacco use, alcohol 70 use, sexual activity) especially among adolescents generally involve huge non-response, or, 71 responses which are not true. Intuitively, disclosure of identity or embarrassment due to 72 unveiling of identity may be main concerns regarding non response or false response. Such 73 responses may cause mainly two problems: first, systematically biased estimate of the burden of 74 tobacco use and alcohol use; and second, distortion in the association between covariates and the 75 use of tobacco/alcohol [2].

76
To overcome above problems, Warner [3] suggested a randomized response technique for 77 estimating the proportion of responses on sensitive issues like tobacco use and alcohol use. As 78 such, the randomized response technique (RRT) provides privacy to respondents by virtually 79 using a randomized device to ask the answer of the sensitive question from them. In addition, 80 this random device plays a pivotal role in concealing the subject's response; and also avoiding 81 the possibility of hiding true response on sensitive issues and also non-cooperation from 82 respondents. Further, according to Diekmann Hglinger, the considered random device needs to 83 have easiness to use, availability, friendly to handle, trustworthiness, and involvement of less large survey on such sensitive topics, it is worthwhile to explore innovative random devices 87 under RRT which may be easier to use and understandable by both, respondents and surveyors.

88
The tobacco and alcohol use among adolescents in India is still considered as a transgression.

89
Thus, this study aimed to explore user friendly random devices for practical utility in large scale 90 field survey while comparing randomized response technique with direct interview method 91 especially on sensitive topics like tobacco use and alcohol use among adolescents. use of a RRT involving an unrelated question whose theoretical frame work was given by 97 Greenberg [9]. Under this design, the respondents are asked to answer if they belong to a specific 98 group which is unrelated to motive of survey. This survey technique conventionally contains two 99 types of questions, one related to sensitive attribute, e.g., "Did you ever use tobacco?", and other 100 unrelated question, e.g., "Were you born in January or February?" Further, a random device is 101 used in such survey to decide about type of questions to be answered by respondents; and thus 102 helps in concealing the subject's response and privacy. So, it plays pivotal role in avoiding the 103 possibility of hiding true response on sensitive issues and also non-cooperation from 104 respondents. This method can be done in two conditions, first when prevalence of characteristics 105 of unrelated question is unknown, and second, when prevalence is known. However, it is also 106 reported that efficiency of RRT involving known prevalence of unrelated question in the 107 population is better than that in case of unknown prevalence[9,10]. They found this approach 108 better than even Warner's randomized response technique. Further, random device used under 109 RRT, should be easy to use, friendly to handle and appropriate for time-bound application.

110
The prevalence of tobacco use under RRT [9] with known prevalence of unrelated 111 question may be estimated as: Where, under standard normal distribution at 95% confidence level is 1.96.

126
Moors [11] reported that randomized response technique had better efficiency if ratio of sensitive 127 question vs. unrelated question is 7:3. So keeping this in view, proportion of students responding 128 question on ever tobacco use i.e., proportion of students who answered questionnaire related to 129 ever tobacco use was taken as 0.70. Likewise, in case of ever alcohol use also, this proportion 130 remained same.

131
Direct Interview survey method is a method where interviewer enquires the required 132 information directly from interviewee. If D is the number of students who had given yes for ever 133 using tobacco in direct interview among n respondents, then the proportion of tobacco user p 134 (TDI) is defined as: For larger n, confidence interval (C.I.) of estimated proportion is given as: Where, under standard normal distribution at 95% confidence level is 1 The primary data was collected though cross-sectional survey exclusively developed for this 160 study. Survey was conducted among students of 9 th to 12 th class from purposively selected three   In comparison to RRT, the direct interview method may provide underreporting of sensitive 300 events which may be pushed further higher side in case of highly sensitive events. The present 301 study has amply shown that assuring privacy of information through use of a random device 302 under RRT will result into more accurate reporting of socially unacceptable behaviours like ever 303 tobacco use and ever alcohol use among adolescents. Further, the findings under the present 304 study has confirmed the utility of the randomized response technique as a versatile method for 305 gaining access to more accurate information on sensitive social problems. As such, the present 306 study has shown the path regarding possible use of such approach (RRT) in estimating prevailing 307 burden in relation to sensitive topics more accurately. This may however require further 308 exploration regarding feasibility of its use under large scale surveys on the sensitive topics.