Experimental Design Details
This study will consist of a single survey that will take between 15 and 45 minutes. A primary focus of the survey is to measure beliefs about the mortality risk of taking motorcycle taxis and the effectiveness of motorcycle helmets at preventing death. The study is a lab-in-the-field experiment in which respondents will receive randomly varying information about the safety risks of motorcycle taxis and helmet effectiveness. They will then participate in a Becker–DeGroot–Marschak (BDM) willingness to accept game to measure their valuations of motorcycle helmets. The goal of this experiment is to estimate how beliefs about safety affect helmet demand, which in turn can be used to estimate the value of a statistical life (VSL).
Respondents will be recruited at motorcycle taxi stands in Nairobi. Surveyors will interview respondents throughout the day, although traffic is generally higher during morning and evening commutes. We expect that many potential respondents will be time constrained and thus aim to minimize the duration of the survey. Surveyors will visit multiple stands throughout the city in order to reach a broader sample of passengers.
Data collection will consist of 1 to 2 weeks of piloting followed by roughly six weeks of data collection. The goal of the pilot phase is to refine the survey questions and resolve any issues with the survey instrument before beginning data collection. In particular, the processing of measuring beliefs about one's mortality risk is likely to pose challenges. A focus of the pilot is in refining the set of questions used to measure these perceptions and the wording of questions.
Data will be collected using SurveyCTO. Randomization will be conducted in SurveyCTO using the random() function. This function uses the Java randomization algorithm to take a pseudo-random draw from a standard uniform distribution. This study does not stratify randomization since the sample is not known en-ante. Hence, randomization must be conducted in the survey. Independent random draws are used to determine which information treatment group the respondent is assigned to and which price offering the respondent receives in the BDM demand exercise. Since a primary focus of this study is estimating the value of a statistical life, which requires data on individuals mortality beliefs, respondents are assigned to the pure control with a lower probability (10% vs 30%). We plan to offer respondents a cash payment between Kenyan shillings (Ksh) 5 and 600 with uniform probability.
The first information treatment group is a pure control. These respondents will be asked a series of demographic questions and basic information about motorcycle ridership, then proceed to the willingness to accept exercise. Those assigned to the pure control will not be asked any questions about their perceived likelihood of dying in a motorcycle accident, or other questions relating to motorcycle safety. The aim of the pure control group is to provide an estimate of baseline demand for motorcycle helmets among consumers that are not asked to think carefully about safety before being offered a helmet.
The control group and treatment groups will be asked detailed questions about their perceptions about motorcycle safety. Prior to the pilot, we do not plan to offer any information about mortality risk or helmet effectiveness to those in the control group. However, we may present those in the control group with data about the respondent's empirical accident risk depending on responses during the baseline. If respondents have extremely diffuse priors about empirical mortality risk, then noise in the measurement of perceptions about accident risks may dramatically reduce the power of this study. In this case, we may present the control group with empirical accident risk data so that the primary variation generated by the study will enter through beliefs about helmet effectiveness which are well-studied, more heavily publicized, and easier to understand.
Empirical risks are calculated using the 2021 mortality risk per motorcycle trip of motorcycle drivers, since we know that one driver is present during each trip but were not able to obtain high quality estimates about the frequency of trips that involve a passenger, calculated using data from news sources and the Kenyan National Transport and Safety Authority (NTSA). We then use the per-trip risk of an average Kenyan and the respondent's ridership volume to estimate their mortality risk over a 5-year span, the recommended lifespan of the helmet.
We currently plan to present both treatment groups with the empirical mortality risk estimates. However, if respondents are able to articulate well-measured priors that exceed the empirical estimates, we may remove the empirical estimates from the survey instrument. There are two reasons for this. First, the empirical estimates are poorly measured due to data constraints, so we do not want to present respondents with information that may make them less likely to engage in safety precautions if they have well-formed priors. Second, we think it is likely that individuals under-estimate helmet efficacy given extremely low use. Revising beliefs about mortality risk down and helmet effectiveness up would generate an ambiguous change in the perceived mortality risk reduction from a helmet, dramatically reducing power.
Both treatment groups will receive information about helmet effectiveness. The first treatment group will be presented with the results of Liu et al. (2008) which conducts a meta-analysis of studies on helmet effectiveness, predominately from developed contexts, and estimates that helmets reduce mortality risk by 42%. Those in the second treatment group will be presented with the finding from Ouellet and Kasantikul (2006) that helmets in Thailand reduced mortality risk by roughly 70% when properly worn.
After completing the information treatment, all arms will engage in a BDM willingness to accept game in which they are asked to state the smallest cash payment they would prefer to a free helmet. A cash payment will then be drawn at random, and respondents will receive the cash if it is at least as large as their bid and otherwise be given the free helmet.
A lack of credible data on beliefs about helmet effectiveness and helmet demand made conducting power calculations difficult. In addition, the primary goal of the treatment is to generate a strong instrument, not detect a treatment effect. Hence, we plan to survey the maximum number of people possible with the budget available for this pilot. We expect the number of surveys to be around 1,000-2,000. If this sample is sufficient, data collection will end. If not, we will collect more data in a second wave and pool the samples, including wave fixed effects in regressions.