Dear s. Benzaim

Analyzing 100 startups can be sufficient for journal publication in economics, but its adequacy depends on the quality of the data, the significance of the research question, and the methodology employed. If the study addresses a novel or highly relevant topic, uses robust econometric methods, and provides meaningful insights, a sample of 100 may be acceptable, particularly for niche or specialized journals. High-quality, detailed data and strong theoretical grounding can compensate for a smaller sample size. However, larger and more representative datasets are typically expected for empirical research in top-tier journals. To strengthen the study, it is important to transparently acknowledge limitations, emphasize theoretical contributions, and ensure alignment with existing literature.

I hope this helps.

Félix Oscar Socorro Márquez thank you professor

Those startups that you study will represent those startups only unless you have some way to infer to a larger population. That is done either by design-based/probability-of-selection-based methods, or a prediction-based approach or some combined method. It seems you cannot use a randomized design-based method because even if you started out that way, you would have a great deal of nonresponse, enough to ruin your design. You'd need a prediction-based approach or a small design-based approach with a way to use that with a larger nonprobability sample. At any rate you would need a good deal of good quality auxiliary or predictor data. Data for a variable you need may not be available. Inferring from nonprobability samples involves these other data and may not be feasible.

However, you might check into https://cran.r-project.org/package=nonprobsvy.

James R Knaub Thanks a lot dear PR. James for your insightful comment.

The primary challenge for this research in Algeria is the complete absence of an official, comprehensive registry of startups.My attempts to secure such lists from public establishments were largely unsuccessful; one provided a list of only 30 successful startups, which is unrepresentative and unsuitable for an SRS of the entire population (including failed ventures). This reflects a broader systemic issue of limited statistical coordination and data accessibility for researchers in the country. Therefore, a Simple Random Sample (SRS) was fundamentally impossible.

Furthermore, even if an SRS were feasible, the universal challenge of non-response bias would remain, as individuals' willingness to participate (whether due to frustration, excitement, or openness) inherently skews any sample.

Given these severe constraints, my data collection strategy was a pragmatic mitigation effort:

1. Targeted Dissemination (Limited Scope): Leveraging the partial list from the establishment and seeking their help to distribute the survey to their contacts.

2. Direct Outreach: Personally contacting startups from any other small, specific lists I could obtain.

3. Broad Online Voluntary Recruitment: Using widely accessible platforms like Facebook to reach a broader, more diverse set of entrepreneurs, including those potentially missed by official channels.

This combined approach yielded 95 responses and taked one year! While not a probability sample, it represents the most exhaustive effort possible under the circumstances.

The analysis (Kaplan-Meier, Cox regression) will thus focus on understanding survival patterns and predictors within this specific, hard-to-reach cohort. Conclusions regarding the broader Algerian startup ecosystem will be drawn with significant caution, explicitly acknowledging the sampling limitations imposed by the operating environment. This study aims to offer crucial initial insights despite these unavoidable data collection hurdles

You mentioned the nonprobsvy package. In a situation like mine, where true population auxiliary data is scarce to non-existent, are there any practical (even if limited) ways this, or similar techniques, could be used to discuss potential biases or perhaps even very cautiously explore adjustments for the Cox model results? Or is it mainly about transparently reporting limitations?Beyond clearly stating sampling limitations, are there specific analytical points or arguments you'd recommend emphasizing in the discussion section to best frame the inferences drawn from the Cox model, acknowledging its exploratory nature due to the sampling challenges?: Are you aware of any published studies in economics or management journals that have successfully navigated similar, severe sampling limitations in under-researched contexts, which might serve as a good example for reporting and framing?

Unfortunately you cannot infer to the overall population. You cannot say how much the remainder of the population differs without additional relevant information. However, though speculative, that doesn't mean your study cannot stimulate ideas, as long as everyone is clear that the remainder of the universe may differ substantially. Also, it may be instructive to do your study with the 95 cases you found and compare results to what you would have found with only the original 30 cases. Of course, even if there were no bias of any kind, the results could vary due to variance. But it might be interesting to see if there is any obvious bias between those two groups. Actually, perhaps compare the group of 30 to the other 65, assuming that made up your group of 95. Best wishes.

James R Knaub unfortunately, I cannot compare the two groups because the 30 cases represent only successful startups, and I do not know which respondents they correspond to, as the responses were anonymous. I also do not know whether all of them responded or not.

I will wait until the sample reaches 120, and clearly state the sampling method in my paper to remain honest and transparent.

Thank you, Professor.

Best regards.