[pg]

The critical mistake here is a misunderstanding of how probability works. If we know that x% of startups succeed, that doesn't mean that each group of founders starting a startup have x% chance of succeeding.

Some people are orders of magnitude more likely to succeed than others. For those it's a good idea to start a startup. For the rest (a much larger group) it's a bad idea.

[Negitivefrags]

There is no misunderstanding of how probability works here.

100% of people who start businesses believe they can succeed, otherwise they wouldn't start one.

Clearly given that such a high percentage of businesses fail, you are not qualified to judge your own chance at success. Therefore the only time you can know that your percentage chance is higher than the average is when you have a 3rd party that is skilled at evaluating such things that tells you so.

In the absence of such, your chances just fall back to the raw figures that the article gives and so the analysis stands up.

[hluska]

Mind if I jump in with an analogy?

In 2012, the Canadian Olympic team sent 281 athletes to compete at the summer Olympics in London. The World Bank reports that Canada's population is approximately 34,000,000.

Using a raw analysis, you could say that, "A Canadian has a 281/34,000,000 probability of reaching the summer Olympics." That is perfectly valid.

However, you cannot use that same technique to judge individual chances of success. Let's say that I am in a room with Brent Hayden.

Brent Hayden (a swimmer who won a bronze medal) is 6'4 and has a very athletic build. I am 25 pounds overweight, pasty faced from too much time in front of computers and completely void of hand-eye coordination.

Clearly, our individual odds of reaching the Olympics are different. The probability that I will make the summer Olympics is, in actuality, far below 281/34M because I am on the left side of the athletic prowess bell curve. Someone like Brent Hayden's probability of making the summer Olympics in, in actuality, higher than 281/34M because they would fall on the right side of the athletic prowess bell curve.

Those sorts of normal distributions happen in startups as well. Some teams are significantly more suited for the demands of startup lives (just like some couples are significantly better suited for the demands of marriage than others are).

[guylhem]

So what you say with your example is that in a layered sample (say separate the population by heigh or muscle mass), the weighted mean is a better estimator than the arithmetic mean of a sample ?

It is not ignored by statistics. http://en.wikipedia.org/wiki/Weighted_mean

[pg]

Your mistake in turn is to assume that it's impossible for people to judge their own abilities.

Lots of people think they could write a decent novel. Most are wrong. Suppose only .01% actually could. If your argument were correct, JK Rowling should assume her chances of writing a decent novel are .01%. She feels fairly confident that she could, but she has to discount that, because people are often mistaken about such things, and "fall back to the raw figures."

It is possible to know that one is good at something.

[EGreg]

Not only whether one is good at something, but whether they are in a good position to try. If you already have lots of traction, or investors, or smart people around you, then it's a good opportunity.

Having said that, what were the people who funded Color thinking?

[Negitivefrags]

That is not an example of JK Rowling judging her own abilities. Her abilities were judged by the market when she published her books.

It's possible to know one is good at something, it's just not possible to judge your own abilities.

[pg]

I'm not talking about an existing book. I'm talking about the case where she's trying to predict the chance she'll be able to write a new book, before she's started it.

Her previous books should give her confidence that she'll be able to write another. That is one way in which she can judge her own abilities.

[lusr]

What you said is true if you're young and you don't have much data to assess yourself with, but even if you're not pg and don't have a model of success well backed with data, the fact is that if you're 30 or 40 years old there are plenty of objective third parties you will have encountered throughout life who have assessed you for free.

For example, if you've been through a demanding course of tertiary education and performed better than average that says something. Indeed any endeavour that required a fair amount of intelligence, skill, and commitment where you came out above average says something: starting other businesses and having done better than average comes to mind.

In addition to the flaw pg pointed out, the second mistake the article's analysis makes is to ignore the number of opportunities you have; it seems to assume you have a single opportunity. In reality, you can make some less than optimal choices in life and still sometimes do just as fine in the end.

If you're young I don't see what's wrong with gambling a bit: if you're 20 and have zero data about how well you will perform, I believe it's still completely rational to aim high. If you don't get the $100 million exit by 25, you can reconsider your options. After 5 years of aiming high you'll have more than enough data to assess whether it's time to change your aim or to keep going.

[TallboyOne]

I'm no mathematician, or even close, but I don't believe only a person's beliefs translate to their real-world chances of success. A lot has to do with your work ethic, motivation, ability to work long hours, focus, upbringing etc.

I also don't think 100% of people who start businesses believe they can succeed. I'm sure a large percantage are trying to wing it and hope they get lucky.

[Negitivefrags]

I'm not saying that a person's belief translates to their success. In fact I'm saying just the opposite. I would imagine most people starting a business think that their chances are higher than the average person because they are special in some way.

The article is effectively saying "Statistically you are probably not special, so you may want to reconsider taking the risk".

[acgourley]

That's rational. Good luck getting any founder to think that way.

[Negitivefrags]

Well I'm a founder myself and I absolutely did think that we were special and had a much higher chance to succeed than the average person.

Unfortunately at this point it's impossible to know if we were just lucky noobs or actually had some idea what we were doing.

[paraschopra]

There does seem to be some flaw in this argument. Within those people who have orders of magnitude higher probability of succeeding, how do you differentiate? My general point is the it is impossible to assign a probability to an individual's success (since there is only one data point). So for calculating probability, you will have to aggregate many individuals and see how many of them become successful. The OP's understanding of probability is entirely accurate if we talk about general set of people who are starting companies (assuming figures quoted are accurate).

Of course, within a set you can create many more subsets but that doesn't mean original set is wrong.

[peripetylabs]

I think the author correctly understood probability theory.

If a population is modelled by a random variable, each individual probability is unknown. What is known is the continuous probability distribution of the entire population.

Intuitively, this means you cannot know which individuals will succeed or fail, you can only know what proportion will succeed -- regardless of the merits of each individual.

The author correctly calculated mean outcome -- more precisely, expected value:

http://en.wikipedia.org/wiki/Expected_value#Univariate_conti...

[guylhem]

Indeed, and one way to improve that is to separate the random variable that represents the population by different random variables (say male immigrant with at least master level studies, female of less than 30 years, other males, other females) and calculate the success rate of each.

They may have a different probability distribution, even if the population average follow a normal law (central limit theorem)

If you invest is the most achieving group instead of distributing your investment across the population, you will have better returns.

[rdl]

It's also totally meaningless to use the population statistics for "all new small businesses" and try to extrapolate to "scalable tech startups in silicon valley".

I'd have probably 50-80% odds of building a successful consulting business or other small technology business which scales directly on labor, generates ~$100k/yr in profit per employee (over salary), up to a few employees, etc.

I'd have lower odds of building a hugely successful consumer startup like Facebook. I'd hope it's a touch higher than for a random farmer in rural China, but still not that great.

I think I have better odds than many for building something in between -- a business focused infrastructure/security startup, maybe not a 1b user $100b business like Facebook, but also not a consultancy. Plenty of $100mm/yr revenue opportunities in the b2b space.

Even if the expected return for each type of business were the same over all people, I probably have a comparative advantage in b2b, and other people may have a comparative advantage in consultancy or in huge consumer startups.

[guynamedloren]

And how would one know they are orders of magnitude more likely to succeed than others?

[pg]

I've talked about this in several essays. Someday soon I'll write one specifically about that topic.

Curiously enough, one of the best ways would be to apply to YC. We have a huge amount of data about which founders succeed, and we work very hard to identify the probable successes among the applicants.

[guynamedloren]

Looking forward to the essay.

Slightly off topic, but how does YC feel about applicants that are currently employed by YC-backed companies? Somewhat conflicting, no?

[EGreg]

pg, not only are you right about this, but there is a greater mistake in the assumptions here -- namely that there can only be two outcomes: $100M with 0.002% chance, and $0 with a 9.998% chance. In reality, there is so much in between, which raises the expected value quite a bit.