Experimental Design Details
Equity split Arrangement Game/Exercises and Matching Outcomes
The treatments will be short video lectures presenting one of two perspectives on equity splits in startups. The equal-splits treatment will feature a video advocating the “1/N rule”—that each founder should receive an equal stake in the company and its profits. Students will be provided with the arguments frequently cited for this division rule: that equity is an incentive so small stakes should be avoided, that uncertainty about future roles and productivity make it hard to predict appropriate shares, and that unequal splits go against basic notions of fairness and can sour your relationship with your cofounder from the beginning.
The unequal-splits treatment will feature a video advocating a formal pie slicing approach—that the founders should give a numerical score to their value added along several dimensions—initial idea, business plan development, domain expertise, commitment & risk, and day-to-day responsibilities—assign weights to each dimension, and use the averaged scores as a starting point to discuss an equity split. Students will be provided with the arguments frequently cited for this division rule: that effort and value added will rarely be equal, so equal splits can create friction as asymmetries become clear; that founders often bring capital to the table or work without pay while the other founder keeps their day job, rationalizing a greater share.
The unequal-splits treatment video will illustrate an excel implementation of the pie slicing activity, which the students in this treatment will perform with their partner later in the program.
Because classes within the same polytechnic will have different treatments, we will acknowledge that the lesson the students receive represents one school of thought, and that other classes may be presented with different rules.
We will follow a direct difference-in-difference approach, comparing treatment and control group at baseline and a follow up. The treatment will last one semester, and we will conduct a follow up one year after the program start.
We decided to follow this approach given our relatively small number of observations in total, and more importantly, our small number of participants institutions --- vocational colleges, community colleges--- in the program, which undermine our statistical power. Standard error will be clustered at institution level.
As outcomes we will use the outcomes specified previously under II. Skills and Decision-Making and III. Entrepreneurship Outcomes. As control we will use all the dimensions specified in the demographic section.
We will correct our p-values by index using family-wise error correction.
Since we care about mechanisms, we are not correcting for multiple testing between all different types of outcomes (Haushofer and Shapiro, 2017).
We will consider a 10% of attrition in our sample. For attrition we will run the following estimations:
1-. Mean test between control and treatment who “attrited” at baseline.
2-. If we add new subject at follow up, mean test between new control and treatment subjects at follow up.
3-. If the attrition is unbalanced between groups, we will also compare the mean differences among non-attritors at baseline.
In order to limit noise caused by variables with minimal variation, questions for which 95 percent of observations have the same value within the relevant sample will be omitted from the analysis and will not be included in any indicators or hypothesis tests. In the event that omission decisions result in the exclusion of all constituent variables for an indicator, the indicator will be not be calculated.
Random allocation of students to different interventions, and their match to random comparison student’s in control colleges, provides an exogenous variation in treatment status that allow us to estimate effects on relevant outcomes previously describe.
To estimate the impact of each of the interventions, we can use the following OLS specification:
Y_ij=α+β_1 T1_ij+β_2 T2_ij+X_ij+S_j+ϵ_ij
where Y_ij is the measure of an outcome of interest post-treatment for individual i in stratification block j (college and race), T1 is an indicator variable if college of student i was assigned to RYSE activities, T2 indicates whether college of student i was assigned to Potential Shares exercises.
X_ij is a vector of control variables at student level, S_j is a stratification block fixed effect, and ϵ_ij is the error term, clustered at the college-level. From this specification, coefficients β_k represent the causal effect of being offered the participation in the k intervention and corresponds to the Intention to Treat (ITT) estimate under imperfect compliance.
To analyze heterogeneity by race, we will use the previous specification and include interactions between each treatment arm and race.
Finally, we will estimates heterogeneous effects based also on work-related dimensions and personality (using the big five taxonomy).
In addition to the direct analysis of the experiment, we will use data from the speed dating exercise to infer preferences by fitting a frictionless matching model.
Variation in class size induces variation in market thickness for the matching process, which will allow us to study the effect of more partner options on each outcome of interest by adding class size into the above specification. Generally, we predict that thicker markets should yield more efficient assignments, though the degree to which this is true will depend on the preference structure and potential instabilities endemic to stable roommates’ problems.
Additionally, we will study a specification where variations in class size are interacted with treatment. We hypothesize that equal splits will lead to less efficient assignments, but has an indeterminate effect of incentives within the partnership. We hypothesize that the costs of equal splits relative to unequal splits will be greater for larger markets (class sizes) as situations where a non-equal split is optimal but blocked by a less efficient equal split become more likely.