Length biased sampling: what the study cannot prove.
Length biased sampling is the first weakness, because panels catch slow attempts more often than fast ones. Shim and Davidsson weighted short attempts up to correct it. Before that correction, the same data gives medians of twenty and twenty-four months.
How length biased sampling fills a panel with slow cases.
Panels find people in the middle of trying a business. A slow attempt sits there year after year, available to be caught. Meanwhile, a fast one is over in a single season. As a result, length biased sampling fills the sample quietly with slow cases. The correction weights short attempts up, so rejecting that choice means losing the three months.
Memory, missing cases and an opposing view.
Also, the dates are memories from interviews, and recall bias in surveys makes them imperfect. Nobody remembers cleanly the month they bought a domain. Next, the sample holds only finished attempts, and the authors say their method does not handle that. Finally, earlier researchers argued that staying in an uncertain venture is rational learning. Shim and Davidsson say their result casts doubt on that, but not that it settles it.
The narrower finding that survives.
Naming these limits of entrepreneurship research matters, because swapping one unchecked belief for another is the same trick. So the surviving version is narrower than anything on a poster. Across 1,673 resolved attempts in four countries, businesses tended to appear early and abandonment came late. However, that is a pattern in a population, not a prediction about you.
Weigh length biased sampling against your own number.
Under your boxed number, write that the study describes a population, not you.
Note which of your dates came from memory and which came from receipts or files.
From the book.
So the version that survives is narrower than the one that would fit on a poster. Across 1,673 resolved attempts in four countries, businesses tended to appear early and attempts tended to be abandoned late. That's a pattern in a population. Nothing in that paper predicts what happens in your case, and I'll keep saying so every time a number in this book tempts you to read it as a sentence handed down about your own life.
I Want to Start a Business but I Never Actually Start, chapter 2: The three-month window
Questions about length biased sampling.
Why does length biased sampling matter for the three-month figure?
Without the correction for slow cases, medians are twenty and twenty-four months. So the three months depends on accepting one statistical choice.
Are the dates in the study reliable?
They are memories given in interviews, so they are not perfectly clean. Nobody recalls the month they bought a domain exactly.
Does the study predict what will happen to my idea?
No, it describes a pattern across 1,673 resolved attempts. Nothing in the paper predicts your individual case.
Read the whole chapter.
This lesson covers one section of chapter 2, “The three-month window”. The book goes further, with the full evidence, the stories and every exercise.
Lessons in this chapter.
- 01 When did my business idea start: date the first act.
- 02 Three month window startup: what the study measured.
- 03 Failed ventures take longer: why stopping is so hard.
- 04 Business idea timeline exercise: count your months.
- 05 Clinkle startup story: three years, no public product.
- 06 Buffer pricing page experiment: strangers and a price.
- 07 Length biased sampling: what the study cannot prove.
- 08 Busywork instead of real progress: five dating traps.