x

We are happy to announce that all trial registrations will now be issued DOIs (digital object identifiers). For more information, see here.
The Limits of Startup Strategy
Last registered on August 15, 2019

Pre-Trial

Trial Information
General Information
Title
The Limits of Startup Strategy
RCT ID
AEARCTR-0004302
Initial registration date
July 02, 2019
Last updated
August 15, 2019 7:36 PM EDT
Location(s)
Region
Primary Investigator
Affiliation
Columbia Business School
Other Primary Investigator(s)
PI Affiliation
Columbia Business School
Additional Trial Information
Status
In development
Start date
2019-07-16
End date
2019-09-30
Secondary IDs
Abstract
In contrast to the traditional model of a firm, a startup is fundamentally inseparable from its founder. This experiment aims to demonstrate that the strategic decisions of startups are determined by a tradeoff between the conflicting professional and personal motivations of their founders. We use priming to explore how the relative importance of entrepreneurial identity affects the behavior of founders who face a strategic choice that will increase startup growth at a personal cost.
External Link(s)
Registration Citation
Citation
Gong, Sara and Jorge Guzman. 2019. "The Limits of Startup Strategy." AEA RCT Registry. August 15. https://doi.org/10.1257/rct.4302-3.0.
Former Citation
Gong, Sara and Jorge Guzman. 2019. "The Limits of Startup Strategy." AEA RCT Registry. August 15. https://www.socialscienceregistry.org/trials/4302/history/51813.
Sponsors & Partners

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
Experimental Details
Interventions
Intervention(s)
This study examines how the profit-maximizing strategic decisions of growth-driven startups are constrained by the personal motivations of their utility-maximizing founders. In a novel experimental approach integrating economics, psychology, and strategy, we apply priming techniques to study entrepreneurial decision-making. We prime the "entrepreneurial identity" of startup founders and then study how a variation in identity salience influences their willingness to make certain choices in hypothetical scenarios.
Intervention Start Date
2019-07-16
Intervention End Date
2019-09-09
Primary Outcomes
Primary Outcomes (end points)
There are two key outcome variables of interest: entrepreneurs' score on the word-choice test describing their entrepreneurial identity dominance (entrep_score), and entrepreneurs' stated willingness to take hypothetical strategic decisions (strat_score). We measure entrep_score by calculating the number of entrepreneurial words a subject selects from the ten word pairs, divided by ten, so that entrep_score is between 0 and 1. We measure strat_score as the average of a subject's responses to the hypothetical scenario questions, where each scenario is a 5-point Likert item, with possible responses as "Very Unlikely" (1), "Unlikely" (2), "Neutral" (3), "Likely" (4), and "Very Likely" (5).

In our analysis, we study these variables in several ways:

(a) First, we run simple regressions showing whether treatment (a binary variable representing those that are primed) predicts a higher entrep_score and/or strat_score.

(b) We also study the distribution of these effects by using machine learning methods that allow us to recover the shape of the effects based on observables. In particular, we apply the honest tree method of Athey and Wagner (2015) and the sorted effects method of Chernozukhov et al (2018).

(c) Finally, we study how the difference in strat_score is determined by differences in the underlying characteristics of subjects. To do so, we run interaction models where we interact our treatment with pre-treatment characteristics and then estimate flexible functions of pre-treatment characteristics with the estimated effects from (b).
Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
We perform the analysis described in "Primary Outcomes" independently on each hypothetical scenario. We do not have any priors at this moment on how their shape or magnitudes will be different.

We also add the following analyses:
(a) Changing our main outcome variable to a binary measure that is equal to 1 if the subjects choose "Likely" or "Very Likely" and 0 otherwise to study LPM, and logit specifications on the odds (or probability change) of being likely choose.
(b) The impact of being more 'entrepreneurial' (as defined by entrep_score) on strat_score as well as the impact of treatment across this distribution.
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
This study examines how the profit-maximizing strategic decisions of growth-driven startups are constrained by the personal motivations of their utility-maximizing founders. In a novel experimental approach integrating economics, psychology, and strategy, we apply priming techniques to study entrepreneurial decision-making. We prime the "entrepreneurial identity" of startup founders and then study how a variation in identity salience influences their willingness to make certain choices in hypothetical scenarios.
Experimental Design Details
Research in psychology and economics posits that individuals have multiple identities centered around the different social roles they occupy, and that their behavior at a given time may depend on which identity is dominant (Akerlof and Kranton, 2000). Priming interventions can cause a particular identity to become more salient, altering an individual's values and desires and, thus, his or her economic decisions (Benjamin et al, 2010). This study examines how the profit-maximizing strategic decisions of startups are constrained by the personal motivations of their utility-maximizing founders. We develop a novel experimental approach that applies priming techniques to study entrepreneurial decision-making and, in turn, how identity limits startup strategy. Our experiment primes entrepreneurs' "entrepreneurial identity" through a survey mechanism, where entrepreneurs are randomly distributed one of two versions of a survey, priming or control. The priming version asks subjects four questions about their startup strategy (e.g., "What is your exit strategy?"), while the control version asks four questions about their work-life balance (e.g., "Whom do you typically spend leisure time with?"). In both surveys, the priming or control questions are followed by questions about how the subject would behave in the following hypothetical scenarios. We designed these questions to specify a trade-off between startup growth and personal utility: A. A major Silicon Valley venture capital firm has contacted you to express interest in investing in your startup. The firm has a history of partnerships with numerous highly-successful companies, and Series A investments in its partners are typically around $5 million. However, the firm only invests in startups that are located in Silicon Valley. How likely would you be to decide to move to Silicon Valley? B. For the last month, you have been employing a close relative to do part-time work for your startup. However, the quality of their work is below expectations, often requiring you to spend extra time correcting their errors. You have spoken to the relative regarding your concerns, but the errors still continue. How likely would you be to fire your relative? C. You have planned a long vacation to spend some quality time with your family and/or your closest friends, but you have just been invited to a special event where you will have the opportunity to network with numerous potential investors. This event is going to be held during the time you had planned to be on vacation. How likely would you be to cancel your vacation? D. Your startup has the opportunity to make an investment that may greatly increase growth in the future. In order to finance the investment, you would have to go into personal debt. How likely would you be to make the investment? The questionnaire also includes a word-choice test to measure the dominance of subjects' entrepreneurial identities. Subjects are asked to choose the word that best describes them from a series of ten word pairs. In each pair, one word is entrepreneurial (i.e., "innovative") and one is not (i.e., "generous"). We study variation in the number of entrepreneurial words chosen across treatment conditions as a manipulation check. Finally, the questionnaire asks respondents the minimum value of their startup to measure whether priming affects startup valuation as well as personal preferences. Our main sample is drawn from a population of entrepreneurs who have registered for a free online class that will be taught by one of us (Guzman). Entrepreneurs register by completing a 10-minute registration form with questions about their personal and startup backgrounds. Several weeks after registering for the class, subjects are asked to fill out a pre-class survey, which constitutes the experiment. We screen our registrants to select young growth-driven startups, selecting only those who answer "Yes" to the question "Are you a startup founder?". We have already conducted a two-phase pilot through Qualtrics Online Samples, and we are now updating our trial registration to include improvements made after this pilot. Our first pilot phase was a soft launch of our surveys on a sample of 12 self-identified entrepreneurs, and our second phase was a full launch on a sample of 100. In our initial trial registration, we had hoped to present the second phase as an experiment in its own right. As of this update after pilot completion, we have found strong evidence for the effect of priming on entrepreneurial identity and strategic decision-making, and have attached our data and R code for analysis. However, we are concerned about the quality of our data: our survey respondents recruited by Qualtrics included, for instance, a contract driver and a teenager who mows lawns. In addition, although survey distribution was randomized, we also found evidence that entrepreneurs with less valuable companies were less likely to complete our priming survey. We also did not collect statistics on attrition and survey drop-out. Therefore, we believe this first experiment does not meet scientific standards, and we intend to present these results not in our main paper but in an appendix. Nevertheless, the results of the pilot provide strong motivation for launching our experiment on our main sample, and were also useful for power calculations.
Randomization Method
We performed block randomization according to the background characteristics of subjects collected at registration. We partitioned the covariate space using the following variables: startup financing stage, startup value, startup location, satisfaction with local resources, satisfaction with local quality of life, number of founders, number of employees, having children, marital status, having a second job, age, and gender. We then grouped similar subjects together, in backwards order of the listed variables, until each block had at least 4 subjects, as recommended by Athey and Imbens (2016), resulting in 18 total blocks. Finally, we performed a balance test (attached) to ensure that treatment and control group means were not significantly different among every covariate.
Randomization Unit
Individual.
Was the treatment clustered?
No
Experiment Characteristics
Sample size: planned number of clusters
The experiment is run at the individual level and not clustered.
Sample size: planned number of observations
We recruited class participants over the course of three weeks. Our total enrollment was 278, and after dropping non-entrepreneurs and minors, we had a subject pool of self-identified entrepreneurs with a startup, to whom we will send out the priming and control surveys. We been able to recruit 121 entrepreneurs for our experiment.
Sample size (or number of clusters) by treatment arms
61 treated, 60 control.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
There is no established value in the literature that can help guide our analysis, so we use the results of our pilot study for power calculations. However, because the effect sizes may have been biased by issues with the quality of our Qualtrics data, we intend to recruit larger samples than suggested by these power calculations. In our pilot study, we found that the mean entrep_score was 0.45 (0.18 SD) for the treatment group and 0.36 (0.19 SD) for the control group. This implied a Cohen's d effect size of 0.51, and thus the minimum sample size for detection (at a significance level of 0.05 with a power of 0.8) was 62 per group. We also found that the mean strat_score was 3.44 (0.81 SD) for the treatment group and 3.14 (0.81 SD) for the control group. This implied a Cohen's d effect size of 0.37, and thus the minimum sample size for detection (at a significance level of 0.05 with a power of 0.8) was 119 per group. We have attached the R code and output for these calculations.
Supporting Documents and Materials

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
IRB
INSTITUTIONAL REVIEW BOARDS (IRBs)
IRB Name
Columbia University IRB
IRB Approval Date
2019-07-02
IRB Approval Number
AAAS5166
Post-Trial
Post Trial Information
Study Withdrawal
Intervention
Is the intervention completed?
No
Is data collection complete?
Data Publication
Data Publication
Is public data available?
No
Program Files
Program Files
Reports and Papers
Preliminary Reports
Relevant Papers