Ignoring Good Advice

In the main wave of the study (wave 2), subjects will undertake two (incentivized) tasks; a random (coin-flipping) task and a real-effort (number-counting) task, which provide within-subject variation. Subjects will then be offered the chance to take advice from an adviser (another subject drawn from wave 1 of the study which we will run in advance of wave 2). We vary the remuneration of the adviser, which provides randomized between-subject variation.

2020-07-30

2020-08-31

binary variable of whether (good) advice followed or not, and its interactions with: adviser remuneration; task type (random or real-effort); task performance; subject envy; susceptibility to the sunk cost effect; (general) stubbornness.

Each subject decides whether to accept or ignore advice, once for each task. Advice is said to be good if accepting it increases expected payoff. We vary the remuneration of advisers, which provides randomized between-subject variation. We will also measure the role of the task itself (luck or effort-based) by looking at the behavior of the same subjects across tasks; a within-subject measure. Subject envy is a measure of envy drawn from the psychology literature. We will use a new measure of susceptibility to the sunk cost effect (the "SCE-8") taken from one of our other papers ("Evaluating the Sunk Cost Effect" by Ronayne, Sgroi & Tuckwell, CAGE WP 475, 2020). General stubbornness is measured using a series of questions taken from the existing literature converted to a Likert scale.

intelligence test; questions relating to medical advice (linked to the COVID19 pandemic); demographics

We will use two questions related to advice-taking behavior in the COVID19 pandemic together to measure recent real-life advice taking. We will use a short (Raven's progressive matrix) test to provide a control for intelligence, and standard demographics to add further controls for our analysis.

WAVE 1: In wave 1, subjects will be recruited using Amazon's Mechanical Turk (MTurk). We will ask them to complete 2 tasks. In one they guess the result of 5 coin tosses. In the second they count the number of 1s in 5 matrices of 0s and 1s. Both tasks will be incentivized. We will reveal their performance in the 2 tasks and ask if they are willing to advise later subjects (advisees; those in wave 2 of the experiment) to submit their answers instead of those chosen by the wave 2 subject. We will select some of the consenting wave 1 subjects as advisers for wave 2. Each selected wave 1 subject will be randomized into a different treatment in which they are paid differently. In one treatment the adviser will receive a payment at the same level as the subjects in wave 2. In a second treatment they will receive a larger payment. A third treatment involves a payment that is only awarded if wave 2 subjects follow their advice, which allows wave 2 subjects to directly influence the remuneration of the adviser. Note that in each case wave 1 subjects will receive the same bonus (incentive) payment instructions as seen by wave 2 subjects, which specify the minimum bonus they will receive: earning a larger bonus (for instance in the second treatment) will not influence the adviser's behavior since this will only become apparent after they have completed the task.

WAVE 2: In wave 2, subjects (different MTurk subjects) will take the same 2 tasks as in wave 1 (which will provide within-subject variation). This time, after taking the tasks, they will be offered the chance to follow the advice of one of the wave 1 advisers (the between-subject treatment). When making this choice, they will also be told: (1) their own performance; (2) an adviser's performance and (3) how the adviser was remunerated. Note that in wave 2 it will be made clear to subjects that the adviser received the same instructions as they did regarding payment, and hence the adviser did not know the size of the bonus before completing the tasks and so this would not have influenced the adviser's behavior more than it did their own. The different levels of adviser remuneration provide between-subject randomized variation since each wave 2 subject is randomly allocated to one of the three wave 1 remuneration treatments for advisers. All wave 2 subjects will then complete a Raven's progressive matrix test, the Dispositional Envy Scale (Smith, Parrott, Diener, Hoyle & Kim, 1999), a test of general stubbornness (Wilkins, 2015), the SCE-8 scale (Ronayne, Sgroi & Tuckwell, 2020) and some general demographic questions as well as two questions on advice-taking in the COVID19 pandemic.

The randomization is undertaken by the software Qualtrics.

Individual level

No

No clustering

25-75 in wave 1; 1500-1800 in wave 2.

500-600 for each of the three (wave 2) adviser-remuneration treatments

Measure: The absolute difference in two proportions (of subjects who take good advice). Our budget is fixed but our payment per subject can vary significantly depending on subjects' choices. Pilots suggest the cost per subject is such that a total of 1500-1800 subjects can be recruited for wave 2. That gives 500-600 subjects per between-subject treatment. The minimum detectable difference in two proportions - regardless of the absolute values of the two proportions - is 0.0885 found in the case that we collect data from 500 subjects per condition. (This is found in the case that the midpoint of the who proportions is 0.5; if the proportions have a different midpoint, smaller differences are detectable.) If we collect 600 per condition, the analogous figure is 0.0808.