Measuring social and economic preferences of the aging rural population in China

Incentive and incentive removal in public goods games centered around elderly care services provision.

2023-03-01

2023-04-30

1. PGG outcome family;
2. Support for the elderly careservices family.

1. PGG outcome family:
1.1 Contribution amount;
1.2 Beliefs about others' contribution;
1.3 Personal norms of others' contributions.
Standardized Index among cases for the same round.

2. Support for the elderly care center family:
2.1 The volunteering days of one's own;
2.2 Speculation of other fellow villagers' volunteering days.
2.3 Personal norms of others' volunteering days.
Standardized Index among cases for the same round.

1. Preferences for individual vs. collective ways to build public goods.

Besides the secondary outcomes, we will also measure a series of variables to explore the possible mechanisms. The variables are:
1. Trust family;
2. Filial Piety norms;
3. Pathway preferences for solutions.

1. Preferences for individual vs. collective ways to build public goods:
In the Public Goods Game scenarios, the participants are asked to first choose their preferred way to pool their resources together to build the elderly care public goods. The first way is to pool resources with members within their social circle, e.g., relatives and friends. The second way is to pool resources with villagers in their village. The former represents a preference for individual solutions and the latter represents a preference for collective solutions.

Besides the secondary outcomes, we will also measure a series of explorative variables to explore the possible mechanisms. The variables are:
1. Trust family;
1.1 General trust. We will use two items to measure the participants' general trust on a 5-point Likert scale.
1.2 Specific trust. We will use six items to measure the participants' trust in specific types of people in their lives on a 5-point Likert scale. The planned six categories are family, relatives & friends, neighbors, fellow villagers, village leaders, and hired help (in terms of elderly care).
1.3 Trust in PGG. We will use four items to measure the participants' trust in other participants' behaviors in the PGG scenarios on a 5-point Likert scale.

2. Filial Piety norms;
2.1 Participants' norms: We adopt and adapt the Three Dimension Filial Piety Scales (Shi & Wang, 2019) to measure this variable. In specific, we use eight items from the dimensions of Family Role Norms and Good Affection. In addition, we adopt two items regarding gender & elderly care responsibilities based on the local input. The items are measured by a 6-point bipolar Likert scale.
2.2 Speculation of others' norms: The same items as above are used to measure the participants' speculation about others' choices. The items are measured by a 6-point bipolar Likert scale.

3. Pathway preferences for solutions.
There will be three questions measuring such preferences.
First, the participants are asked to rank their preferences on five types of solutions for providing elderly care to their parents. Each type of solution indicates either an individualist pathway or a collectivist pathway.
Second, the participants are asked to rank their preferences on five types of solutions for providing elderly care to themselves in the future. Each type of solution indicates either an individualist pathway or a collectivist pathway.
Third, the participants will be given a scenario where there are four ways to use an elderly care subsidy. The participants are asked to choose among the four, one is to receive it by individual account and use the subsidies on one's own; the second is to receive it by individual account and pool the subsidies together with friends and relatives; the other two ways are to receive it by public account and use the subsidies collectively.

The study uses a randomized control design at the villager level.

It is a between-subject treatment in terms of incentive strategies, where one-third of the participants are assigned to the control group, one-third are assigned to the monetary incentive group, and the rest one-third are assigned to the norm appeals group. In each condition, the participants play two rounds of public goods games on building a public good regarding elderly care. The difference between conditions is the introduction of incentives. We will introduce the incentives in the first round, then remove the incentives in the second round.

Model:
For the whole sample, the statistical model will be a linear regression for the primary outcomes to estimate differences in means (intent-to-treat). We will introduce dummy variables to control the confounding effects of rounds. The individual fixed effect and case fixed effect will be included in the model. Robust standard errors will be presented.

PGG_outcome_ir= alpha + beta_1 money incentive +beta_2 norm appeal + round_dummies + individual_fe + case_fe + epsilon

In addition, we intend to conduct sub-sample analyses for the secondary outcomes and explorative variables to explore the mechanism.

Attrition:
Since the intervention is embedded in the survey, we expect very little attrition. A possible scenario is some participants may drop out in the middle of the survey. The balance of missing outcome data will be presented by treatment status. If balanced by treatment status and if attrition is small (<10%), no adjustments will be made. If missingness is systematic and substantial, we will use bounds (such as Lee Bounds).

Heterogeneity:
Subgroup analysis will include separate linear regressions by males and females. Since we will collect auxiliary data on the potential mechanisms (e.g., trust, norms), we expect to be able to control for heterogeneity.

Spill-overs:
1) The experimenter will ask the participating villagers not to discuss the content with each other after the session. Given the gravity of the topic, the villagers are unlikely to break the rules.
2) The participating villagers' residences scatter in the village and there will be very few occasions for them to discuss the experiment.
3) Since the survey contains many questions, the villagers are unlikely to mention the questions in full detail to others, let alone discern the differences between conditions.

Multiple hypothesis testing:
We will calculate p-values and false discovery rate corrected q-values (Benjamini & Hochberg,1995) by families of indicators.

Average standardized effects:
We will give average effects on standardized indicators (Kling, Liebman, and Katz 2007) by families of indicators.

Randomization will be done by the random alternation of questionnaires when they are distributed.

Individual

No

We will enroll 6-7 villages to participate but the randomization happens at the individual level.

540 participants in total 540 participants*2 cases=1080 observations

180 participants in condition 1 (control), and 180 participants in condition 2 (monetary incentive), 180 participants in condition 3 (norm appeal).

The sample size was determined based on power calculation for the test of differences in proportions. We adopt the data from one similar study (Chen, Pillutla, & Yao, 2009). We estimate the sample size for the regression model for all samples by the “sampsi” command in Stata. The previous study shows that in the first round, in the control condition, the individual's contribution portion is 45.8%, in the monetary incentive condition, the individual's contribution portion is increased to 61.2%, and in the appeal condition, the contribution portion is increased to 62.8%. In the second round, in the control condition, the individual's contribution portion is 30.3%, in the monetary incentive condition, the individual's contribution portion is decreased to 35.0% due to removal, and in the appeal condition, the contribution portion decreases to 50.1% due to removal. Among the comparison results, the smallest impact is between the control group (45.8%) and the monetary incentive group (61.2%) in round 1. Thus, we use this minimal detectable impact from the previous study to identify our sample size. Other impacts are all larger than the result of this comparison. In specific, In STATA 16.1 the command [sampsi 0.458 0.612, power(0.8) alpha(0.05)] provided a sample size of 177 per experimental arm. We rounded up to 180 observations. The target total sample size for two treatment arms and one control condition is therefore 540 (3 conditions * 180 villagers). In addition, we can control for individual-level covariates and we double the number of observations by playing the game twice. We can even identify smaller impacts given these conditions.