|
Field
Trial Status
|
Before
completed
|
After
on_going
|
|
Field
Trial End Date
|
Before
June 30, 2024
|
After
June 30, 2026
|
|
Field
Last Published
|
Before
August 11, 2024 12:42 PM
|
After
April 08, 2026 05:08 PM
|
|
Field
Intervention Start Date
|
Before
May 01, 2024
|
After
April 15, 2026
|
|
Field
Intervention End Date
|
Before
June 29, 2024
|
After
June 29, 2026
|
|
Field
Primary Outcomes (End Points)
|
Before
We have two dependent variables. Attention is coded 0 if the mentor does not view the profile and 1 if the mentor views the profile. Connection is coded 0 if the mentor declines the connection request and 1 if the mentor accepts the request.
We will continue to measure the attention variable in study 2 but it is not relevant as we have removed related hypotheses.
Our main variable for all follow-up studies is connection only.
|
After
Connection is coded 0 if the mentor declines the connection request and 1 if the mentor accepts the request.
|
|
Field
Primary Outcomes (Explanation)
|
Before
Potential variables that would be constructed are based on any data we can collect on mentors based on LinkedIn and other archival data
|
After
Potential variables that would be constructed are based on any data we can collect on founders on LinkedIn and other archival data
|
|
Field
Experimental Design (Public)
|
Before
A randomized experiment on requesting LinkedIn connections from entrepreneurship mentors.
In the second study, we will request connections with a personal message instead of just requesting to connect.
|
After
A randomized experiment on requesting LinkedIn connections from entrepreneurs.
|
|
Field
Randomization Method
|
Before
Randomization done in office by a computer using stata. We will upload the log file of the randomization code.
The updated log file for study 2 will be uploaded as well.
|
After
Randomization done in office by a computer using stata. We will upload the log file of the randomization code.
The updated log file for phase 2 will be uploaded as well.
|
|
Field
Randomization Unit
|
Before
We randomize at the mentor level.
|
After
We randomize at the founder level.
|
|
Field
Planned Number of Clusters
|
Before
We use two samples to test our hypotheses. We build Sample A from a national network of entrepreneurship mentors. The network includes 1850 mentors, of whom 1261 have identifiable LinkedIn accounts.
We build Sample B from a second national network of mentors. The Sample B network includes 1090 mentors.
|
After
Phase I YC stratification: gender × followers × pro-bono signal × red state. Phase II stratification: gender × LinkedIn activity level × top university. Pro-bono signal is unavailable in the 2026 scrape. Red-state is excluded from stratification (missing for 31.3% of sample) and retained as a heterogeneity variable. LinkedIn activity replaces follower count as the activity-based variable and is available for 100% of the sample. Eight cells, minimum N = 109.
|
|
Field
Planned Number of Observations
|
Before
840 observations for each analysis comparing two groups (420 mentor assignment per profile).
|
After
Phase I drew its YC sample from LinkedIn search results using self-reported affiliation (n = 2,921). Phase II uses a near-complete census from the official YC directory (raw N = 9,104; after Phase I exclusion and US restriction: 5,488 founders, 3,500 companies; randomized N = 5,476). This is a 1.9× increase over Phase I and eliminates the self-selection bias of the Phase I sampling frame.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
420 mentors for each of the 3 arms (i.e.,3 types of profiles)
Study 2: 545 observations for each treatment group
Study 3: 613 mentors randomized between treatment and control
Study 4: We continue to collect mentor profile data for the Eastern European sample
Study 5: YC sample. 2921 mentors randomized between treatment and control. Randomization is stratified based on gender, number of followers (above to below median), whether they mention probono in their profile and location (red or blue state).
|
After
Eight cells, minimum N = 109.
|
|
Field
Power calculation: Minimum Detectable Effect Size for Main Outcomes
|
Before
|
After
Phase I YC baseline. The full Phase I YC sample (n = 2,686 with controls) produced a treatment coefficient of −0.009 (SE = 0.014, p > 0.10) — a null result — against a White profile acceptance rate of 17.5% (regression constant = 0.175***). The corresponding raw gap was 1.1 pp (White 17.4%, Black 16.4%, p = 0.44). The most relevant Phase I signal for Phase II is the pro-bono mentoring interaction: Treatment × Mentoring = −0.120** to −0.122**, indicating a gap of approximately 11–12 pp among founders who signaled selective engagement. This pattern — discrimination concentrating when founders make deliberate, higher-commitment decisions — is the direct Phase I precedent for H2. Phase II power calculations use the manuscript regression baseline of 17.5% for the simple connection condition and assume a 40–50% reduction (to ~10%) for the advice request condition based on the literature on commitment-level effects.
Minimum detectable effects. With n = 1,369 per arm (α = 0.05, two-tailed, 80% power), MDEs are as follows. For H1, the pooled race main effect (n = 2,738 per race), MDE = 2.52 pp; power to detect a 2 pp gap is 60% and a 3 pp gap is 92%. For H1 (simple connection only, WS vs. BS, baseline ~17.5%), MDE = 4.1 pp two-sided; power to detect a 3 pp gap is 57% and a 5 pp gap is 96%. For H2, the race gap in the advice request condition (WA vs. BA, assumed baseline ~8.5%), MDE = 3.0 pp two-sided (2.65 pp one-sided); power to detect the expected 3 pp gap is 80%. H2 is thus the best-powered individual test in the design, consistent with the theoretical and Phase I pro-bono evidence that discrimination concentrates in selective, higher-commitment contexts.
Power for H3 (DiD/interaction). Under the expected scenario (a 3 pp gap in advice requests and a 1.1 pp gap in simple connections) the expected DiD is 1.9 pp. Two-sided power for this DiD is approximately 19% (28% one-sided). Power reaches 40% at a 3 pp DiD, 63% at 4 pp, and 83% at 5 pp. The study would require a much larger sample to achieve 80% power for the expected 1.9 pp DiD which is not feasible as we used the full YC sample. We address this limitation through three design choices. First, ANCOVA with pre-treatment covariates (followers, activity score, top university, elite employer) reduces residual variance by an estimated R² of 10–18%. This meaningfully improves power for H2 (lowering the advice-request race gap MDE from 3.0 pp to 2.70–2.83 pp and raising H2 power from 80% to 84–87%) but yields only modest gains for H3, raising DiD power from 19% to approximately 22% two-sided. Second, H3 is tested one-sided given the unambiguous directional prediction and Phase I precedent (Treatment × Mentoring = −0.120**, p < 0.01), raising power for the expected 1.9 pp DiD from 19% to 28%. Third, H2 is designated primary and H3 secondary, so the study’s inferential burden falls on H2 where power is adequate.
Informative null. If H2 is null, that result is substantially more informative than Phase I's null. Phase I could rule out gaps larger than approximately 3.6 pp in the YC network (a 20% relative effect at the 17.5% baseline). A null on H2 would rule out gaps larger than 3.0 pp in the advice request condition — a 35% relative effect at the ~8.5% advice-request baseline. Ruling out a 35% relative discrimination effect in a high-stakes context would constitute meaningful evidence that discrimination is not operating, directly addressing the AE's concern that Phase I's low-stakes context precluded conclusions about higher-stakes interactions.
|
|
Field
Intervention (Hidden)
|
Before
1. THEORETICAL AND EMPIRICAL CONTEXT
Black entrepreneurs in the United States are notably disadvantaged relative to their White counterparts. This disadvantage primarily stems from differential access to resources (Bates, Bradford, & Seamans, 2018). Although scholars have closely attended to differentials in the acquisition of financial capital by Black entrepreneurs (e.g., Fairlie, Robb, & Robinson, 2022; Younkin & Kuppuswamy, 2018), less attention has been given to differentials in the acquisition of social capital, or durable networks of social relationships granting access to actual and potential resources (Bourdieu, 1986). However, social capital is an important resource for entrepreneurs (Gedajlovic et al., 2013), and it is a form of capital particularly sensitive to racial dynamics (Putnam, 2007).
To explore the relationship between race and the acquisition of social capital by entrepreneurs, we offer a series of hypotheses tested in the context of LinkedIn, the most used professional social network in the United States. Entrepreneurs used LinkedIn to acquire social capital, such as mentors, potential collaborators, and fellow entrepreneurs. Furthermore, because there is a strong norm for the inclusion of a headshot photograph, race is very salient in the context of LinkedIn.
2. HYPOTHESES
H1: White entrepreneurs will be more successful in attracting the attention of and establishing a connection with potential mentors than are identical Black counterparts.
H2: Racial match between the potential mentor and entrepreneur increases the likelihood of successfully attracting attention and establishing a connection for Black entrepreneurs more than it does for White entrepreneurs.
H3a: Black entrepreneurs who signal activist orientation will be more successful in attracting attention and establishing a connection with Black potential mentors relative to Black entrepreneurs who do not signal activist orientation.
H3b: Black entrepreneurs who signal activist orientation will be less successful in attracting attention and establishing a connection with White potential mentors relative to Black entrepreneurs who do not signal activist orientation.
H4: Black entrepreneurs signaling activist orientation will convert the attention of White potential mentors into a connection at a greater rate than will Black entrepreneurs who do not signal activist orientation.
3. METHODOLOGY
3.1 Design
3.1.1 Constructing the LinkedIn Profiles
We construct identical LinkedIn Profiles for three fictional entrepreneurs. The profile includes a description of a fake tech company and its associated product . The profile also includes educational history and connections. The only differences between the three profiles is in the profile picture. For the activist profile, we add a ribbon to the profile picture declaring #equalitynow for one of the Black profiles. In order to construct verifiable accounts, we used faces and names of the research team and their colleagues. To ensure match on important characteristics we manipulated the three faces using an AI facial manipulation software, generating a total of 25 variations of the three faces. Each variation was independently rated by at least 50 raters recruited via MTurk along the dimensions: attractiveness, trustworthiness, friendliness, competence, intelligence, and age. One of the three faces required two additional rounds of manipulation and rating to achieve a match on the dimensions. The three faces selected were all evaluated as being between 37 and 39 years old with no significant differences on any of the aforementioned dimensions. Thus, we will be able to conclude that any differences generated by the different profiles will be due to the two treatments (entrepreneur race and entrepreneur activist signaling).
3.1.2 Constructing the sample of potential mentors
We use two samples to test our hypotheses. We build Sample A from a national network of entrepreneurship mentors. The network includes 1850 mentors, of whom 1261 have identifiable LinkedIn accounts. We use Sample A to test H1, H3b, and H4. However, due to a lack of racial diversity in the Sample A network, it is not suitable to test H2 and H3a, which both require a substantive presence of Black mentors in the sample. We therefore build Sample B from a second national network of mentors. The Sample B network includes 1193 mentors, and because a LinkedIn account is required by the network, our Sample B comprises 953 mentors.
POWER ANALYSIS – based on pilot study
H1
MDE: White profiles will have a 38% conversion rate of connection requests (this close to 40% which is the average provided by LinkedIn). Black profiles will have a 39% conversion rate of connection requests.
Power: 0.8 alpha 0.05 and Beta 0.02
This is results in a sample of 862
3.1.3 Protocol
The mentors will be randomly assigned to one of three conditions (LinkedIn profiles), two treatments (the Black profile and Activist profile) and a control (the White profile). The randomization for Sample A is stratified on gender, location, and LinkedIn activity. We stratify on the latter because not all mentors appeared to be active LinkedIn users. The randomization for Sample B is stratified on race, gender, and location.
Each mentor will be assigned only one profile. The profile will send out a connection request without an invitation message.
3.2 Variables
We have three independent variables. Entrepreneur Race is coded 0 for the White profile and 1 for the Black profile. Mentor Race is similarly coded 0 for White and 1 for Black. Activist Signal is coded 0 for profiles without the #equalitynow ribbon and 1 for those profiles with the ribbon.
To analyze variation in outcomes, we will use the different variables we have from the scraped LinkedIn and Score data as control/interaction variables. These include but are not limited to profile activity, image features, mentioning SCORE in profile, etc. We have closed the data collection of SCORE mentors on 2/13/2024. The FI data collection will start in late March.
STUDY 2 - New hypotheses and edits
Based on our experience with Study 1, we made several changes to our hypotheses. However, our core interest—differences in the capacity of Black and White entrepreneurs to make social connections—remains the same.
First, we dropped hypotheses relating to the conversion of profile views to contact request acceptances. We dropped this hypothesis because we found that many mentors simply accepted the contact request—which did contain the manipulation—rather than viewing the full profile.
Second, we dropped hypotheses relating to activist choice homophily. Our initial power analysis led us to believe that we just barely had adequate power to test activist choice homophily, with the constraint being the number of Black mentors in our sample. This power analysis was based on our Pilot, which had higher acceptance rates from mentors than Study 1. Based on our response rate from Study 1, we would need a greater number of Black mentors in our sample than we have in Study 2. Additionally, we removed several Black mentors from our sample due to being outliers in the number of connections they have. Study 1 suggests that the number of connections a mentor has is a strong predictor of the likelihood they accept a request from an entrepreneur.
The results of Study 1 as well as the literature on the “underdog thesis” leads us to test a different series of hypotheses in Study 2. The underdog thesis holds that groups which have experienced structural disadvantage are less biased towards or are otherwise more accommodating of marginalized groups (Cech, 2024; Robinson & Bell, 1978).Subsample analysis of our results from Study 1—which demonstrated a bias against Black entrepreneurs at the .1 significance level—provide evidence for the underdog thesis. Chiefly, we find that though male mentors do display evidence of bias against Black entrepreneurs, female mentors do not. Additionally, we find that mentors in states that tend to be politically Republican display evidence of bias against Black entrepreneurs, but that mentors in Democratic states do not. This is consistent with the tendency of Democratic ideologies to be supportive of underdogs. We thus intend to test the following hypotheses in Study 2.
• Female mentors show less racial bias than do male mentors.
• Mentors from underrepresented minorities (African and Hispanic American) show less racial bias than do White mentors.
• Mentors in the bottom third of the network in terms of educational prestige (as measured by the US News and World Report ranking of their alma mater) show less racial bias than do those in the top two thirds.
• Mentors in Blue states show less racial bias than do mentors in Red states.
STUDIES 3 and 4
We found very little evidence of racial bias in our first two studies. Here, we are looking to run the audit study with three other networks. The first two networks are also part of the Founder's Institute. We had restricted our audit study to FI United States. However, there are also FI networks in Europe (England (320 mentors) and Eastern Europe (360 mentors)) and Africa (Nigeria (341), Ghana (164), and Kenya (88) and Zimbabwe(20)) that we include in this study. Our interest in running it in Europe is that audit studies often reveal greater bias in Europe than in the United States. We are also interested in studying racial bias in Africa.
Study 5
We also want to run one study with non-mentor network. Mentors are people who are looking to "give back" and thus may be an easier form of social capital for entrepreneurs from marginalized groups to acquire. We will conduct a final audit study on approximately 2700 alumni who are part of the Y Combinator alumni network on LinkedIn. About 400 of these alumni mention Pro Bono in their LinkedIn profile. We will stratify on this characteristic for balance & heterogeneity analysis.
|
After
|
|
Field
Secondary Outcomes (End Points)
|
Before
I study 2, and follow-up studies we are also sending a message with the connection request. Response to the message from the mentor is a potential secondary outcome to explore. A potential additional outcome would be to use AI and measure the tone or helpfulness of the reply across the two groups.
|
After
In follow-up studies we are also sending a message with the connection request. Response to the message from the mentor is a potential secondary outcome to explore. A potential additional outcome would be to use AI and measure the tone or helpfulness of the reply across the two groups.
|
|
Field
Pi as first author
|
Before
No
|
After
Yes
|