Experimental Design
The design has been revised after piloting: (i) to include a more tempting alternative to studying and (ii) to conduct the experiment online due to the Covid-19 pandemic.
Background
Virtue goods are characterised by unpleasant consumption and long-term benefits. It has been argued that they pose self-control problems that consumers can overcome by buying them on ‘membership’ contracts (i.e., paying a fixed price for unlimited consumption over a given period) rather than ‘pay-as-you-go’ contracts (i.e., paying separately for each unit consumed). It has been found that many gym users choose membership contracts when pay-as-you-go contracts would have been cheaper (‘paying not to go to the gym’). We consider four possible explanations for this finding:
• Self-control (the most commonly cited explanation): membership contracts are chosen as a means of overcoming self-control problems.
• Pain of paying: people prefer to consolidate several small payments into one larger one, and/or prefer to ‘get it over with’ by paying in advance of consumption.
• Prediction error: people over-predict future consumption of virtue goods, and wrongly expect membership contracts to be cheaper.
• Insurance: people are uncertain about their future consumption; because of risk- or ambiguity-aversion, they prefer to pay a known sum of money rather than to be uncertain about how much they will pay.
We aim to discriminate between these hypotheses by using a laboratory experiment in which the virtue good is studying for a test.
Experimental design
In Stage 1 of the experiment, each subject tries each of three computer games (boggle, memory, 2048, each with 10 levels of increasing difficulty) for a short time. She is told that, during the experiment, she will be able to play these games. The games are intended to be a tempting alternative to study. In order to make gaming more entertaining, while playing a game the participant is informed of how her best performance in that game compares with that of other experimental participants.
In Stage 2, the procedures for the remainder of the experiment are explained. Each subject is told that she will be tested on her knowledge of ‘facts’ about an imaginary country and paid according to her test score. She is given an opening balance of money and told about four alternative contracts (described below) for buying access to eight ‘book chapters’ which contain the relevant facts. She ranks the contracts in order of preference. Two contracts are then selected by a random mechanism that gives each possible pair of contracts a positive probability of being selected. The subject is allocated to whichever of these two contracts she ranked higher.
Stage 3 is a 20-minute experimental time that starts with a ‘leisure period’ in which she chooses one of the three games and plays it for at least one minute. After the minute has elapsed, the subject has the opportunity to start a ‘study period’. A subject can initiate a study period at any time, except that each period must be preceded by a leisure period lasting at least one minute. In each study period, the subject views one chapter and can take ‘notes’ about the information it contains. Access to chapters is charged according to the subject’s contract. The subject knows her notes will be downloadable at the end of the session and accessible online during the exam.
In Stage 4, each subject answers a questionnaire about her perceptions of Stages 2 and 3. The questions relating to Stage 2 are designed to elicit attitudes that correspond or conflict with our alternative hypotheses about the criteria individuals use to rank contracts. The questions relating to Stage 3 elicit how far the subject perceived that stage as posing a self-control problem and whether she believes she made errors in predicting how many chapters she would view.
In Stage 5, a few days after the experimental session, each subject takes an online multiple-choice test. The test questions are randomly selected from the facts in the book. Subjects are able to consult their notes during the test. Subjects know they will earn money according to the number of correct answers they give in the test.
In the final Stage 6, each subject is told her earnings from the test, which are credited to her account. She is then paid according to the closing balance in her account. Opening balances and contract terms are set so that closing balances cannot be negative.
Contracts
The four contracts are defined as follows, in terms of parameters x and y, where y > x.
• In the Book contract (a membership contract), the subject buys access to all eight chapters at once before the beginning of the video, making a single payment of £8x, deducted from her balance before the video starts.
• In the Chapter contract (a pay-as-you-go contract), the subject buys access to one chapter at a time, if and when she starts a new study episode. For each chapter viewed, there is a payment of £y, deducted from her balance when the choice to view is made.
• In the Book-rebate contract (BR), the subject buys access to all eight chapters at once before the beginning of the video, making a single payment of £8x, deducted from her balance before the video starts. If the average deduction per chapter viewed is greater than £y, a rebate equal to the difference is added to her balance after the end of the video.
• In the Chapter-rebate contract (CR), the subject buys access to one chapter at a time, if and when she starts a new study episode. For each chapter viewed, there is a payment of £y, deducted from her balance when the choice to view is made. If the total deduction is greater than £8x, a rebate equal to the difference is added to her balance after the end of the video.
Thus, if q is the number of chapters read, Book is less (more) costly than Chapter if q is greater than (less than) 8x/y. BR and CR are exactly equivalent in terms of net payments, and always equal to the cheaper of Book and Chapter.
A major purpose of the experiment is to investigate inconsistencies between subjects’ ex ante rankings of Book relative to Chapter and their choices about how much to study. There is over-commitment (‘paying not to study’) if a subject ranks Book above Chapter, is allocated to Book, and chooses q < 8x/y. There is under-commitment if a subject ranks Chapter above Book, is allocated to Chapter, and chooses q > 8x/y. Intuitively, inconsistencies are most likely to be observed if the values of x and y induce an approximately equal split between subjects who rank Book above Chapter and subjects with the opposite ranking. We will use pilot studies to fine-tune these values to achieve this. Since inconsistencies can be observed only when subjects are allocated to whichever of Book and Chapter they rank higher, we need there to be adequate numbers of such observations. To ensure this, the random mechanism that selects pairs of contracts in Stage 2 will select {Book, Chapter} with greater probability than the other pairs.
Hypotheses and tests
All four hypotheses imply that (after allowing for random ‘error’), over-commitment is more frequent than under-commitment. However, they have different implications about how BR and CR are ranked in relation to each other and to Book and Chapter.
The prediction error and insurance hypotheses assume that the individual maximizes expected utility, given her beliefs about her future utility function. Within this theoretical framework, BR and CR are exactly equivalent and weakly dominate Book and Chapter. Thus, both hypotheses imply that BR and CR are ranked equally or randomly, and that both are ranked above Book and Chapter. In contrast, the other two hypotheses allow ‘anomalous’ rankings of BR and CR.
The self-control hypothesis assumes that the individual has a ‘planning self’ which chooses between contracts and an ‘impulsive self’ which chooses how much to study. Study has greater weight in the preferences of the planning self. The planning self correctly predicts the choices of the impulsive self and treats these as constraints. Since both selves view BR and CR as equivalent, and both prefer low marginal prices for study, BR and CR are ranked equally or randomly relative to each other and above Chapter. But because it can be used as a self-control device, Book may be ranked above BR and CR.
According to the pain-of-paying hypothesis, the individual has a ceteris paribus preference for earlier payments over later payments and for consolidating small payments into large ones. These assumptions imply a strict preference for BR over CR, and weak preferences for BR over Book, and for CR over Chapter. We will use the experimental data to test whether these anomalies occur. Conditional on finding evidence of overall over-commitment, we will interpret these tests as tests of the respective hypotheses.
The prediction error and insurance hypotheses do not predict anomalies in the rankings of BR or CR. But if our data disconfirm the self-control and pain-of-paying hypotheses, the surviving hypotheses gain credibility.
In interpreting the results of the experiment, we will also use the questionnaire data collected in Stage 4. The questions about Stages 2 and 3 provide evidence that can support (or fail to support) each of the hypotheses. In particular, these questions discriminate between the prediction error and insurance hypotheses. They also complement the behavioral tests of the self-control and pain-of-paying hypotheses. For example, Stage 2 responses might provide evidence that many subjects thought about self-control when ranking contracts, and Stage 3 responses might provide evidence that many subjects were conscious of facing self-control problems when choosing whether to study. Such findings would constitute direct and indirect evidence in support of the self-control hypothesis.