Randomization Unit
To ensure the sample is representative of the Japanese household population, we employ stratified randomization (quota sampling) across 10 distinct strata. These strata are defined by gender (male and female) and five age cohorts in 10-year increments, ranging from 20 to 69 years old (i.e., 20s, 30s, 40s, 50s, and 60s). The cell proportions are aligned with the latest Japan Census data.
This is a pure conjoint experiment designed to estimate the marginal utility of specific child attributes on parental educational investment decisions. As such, there are no traditional "Treatment" and "Control" arms at the respondent level. Instead, the "treatment" consists of the randomized variation of child-level attributes (e.g., gender, academic aptitude, or personality traits) presented to each respondent across multiple choice tasks.
In each task, respondents are asked to make decisions regarding Relative Resource Allocation between two hypothetical children with different attribute profiles. By analyzing how these randomized variations influence parental preferences for prioritizing resources (such as tuition and cram school fees) and their Desired Educational Attainment, we identify the causal drivers of educational investment without the need for a separate control group. The within-subject variation of the children's profiles serves as the basis for causal inference.