This article is a crosspost with Harvard Business Review.

Eskimos have 50 words for snow. Humans only use 10 percent of our brains. We hear these types of “facts” all the time — but are they true? Scientists are now saying, “Not so simple.”

We have all seen how repetition of a particular statement or idea tends to lend it legitimacy – the so-called “truth effect.” This effect is likely strengthened when the assertion is made in a serious context by intelligent people with authority. Consider the idea, increasingly an assumption of fact, that “startups create jobs.” Since President Obama exhorted Americans to create startups, and the U.S. government, the Kauffman Foundation, and other partners launched the Startup America Partnership (which launch I attended), startups have been increasingly put forward as drivers of economic growth, in large part because it’s become accepted as fact that startups create jobs. But do they?

Here are five significant challenges to this widely-accepted policy “fact”:

What do you mean by “startup”? The word “startup” has become TV-worthy, with the popular shows Silicon Valley, Shark Tank, and Apprentice purveying passionately pitched startups. But if you probe further, you’ll find that different people mean radically different things when they use the term. Researchers define startups as newly registered firms, sometimes with at least one employee (often the founder). This startup definition is not wrong, obviously, but when economists say “startup,” the vast majority of people think “WhatsApp” and “Snapchat.” Revered Silicon Valley investor Paul Graham says that “a startup is a company designed to grow fast.” Others disagree, arguing that “startups are a state of mind,” “a feeling,” or “a temporary organization designed to search for a repeatable and scalable business model.” This inconsistency is a problem for anyone using the term in an economic policy context.

A related challenge stems from what is called the “survivor bias”: Since so many new companies fail in the first years, the few outliers who do jump successfully through the many hoops are by definition more robust as businesses, and so dramatically distort our views. Hence, you could say “startups create jobs” – as long as you ignore the large majority that don’t.

Not all jobs are created equal. And as far as we know, jobs created by startups – at least the newly-formed-firm variety – are less equal, in two ways. One is that they pay less: A study in the UK has shown that on the average, a startup reaches only $180,000 in revenues after its sixth year, barely enough to pay salaries. A recent study in Denmark has found that indeed, startups create quite a few jobs, but that a disproportionate number of them are low-skilled service jobs.

And keep in mind that even successful startup entrepreneurs work for low or no salary for months or even years after their company appears as a “startup” in census data. Consider this quick story. A few years ago, I met with the winners of the Nordic Startup Awards. While congratulating them, I asked: How many people are employed at your firm? We three just hired our fourth. Me: Interesting, would you mind telling me whether you four are drawing salaries? Oh, no salaries: We are bootstrapping (i.e. working for free to get things going). Me: Are all four of you registered as employees?Yes.

The economic research on “startups” displays a lot of anomalies. Consider the recent World Economic Forum report specifying which countries have the most startup entrepreneurs. Uganda is number 1, in which a remarkable 28.1% of the population are entrepreneurs; Thailand is 2, Brazil 3, and Cameroon number 4. Also consider a recent Kauffman Startup Index showing that the top three startup states in the U.S. are – wait for it – Montana, Wyoming, and North Dakota. Furthermore, the amount of venture capital bears a zero correlation to the amount of startup activity, and is strongly negatively correlated with the presence of mid-market firms ($10 million to $1 billion), which according to a recent Dun and Bradstreet study have created over 90% of jobs since 2008. In the WEF report, the amount of entrepreneurship in a country is negatively correlated with its national competitiveness. A Stanford study reports that startups are also the biggest job shedders. Further confusing the picture, research by the Danish Business Authority shows that among beneficiaries of its programs, 72% of jobs are created by existing firms, 10% by startups, and, incidentally, public investment per startup job created through business support programs costs three times more for startups than for existing firms.

Correlation does not imply causation. Implicit in conventional “startups create jobs” thinking is a logical disjunction leading us (perhaps unintentionally) from descriptive correlation to causative policy prescriptions. Even assuming, for the sake of argument in this discussion, that most new jobs are indeed found in new firms, it still does not necessarily follow that increasing the numbers of new firms will increase the numbers of new jobs. Take, for example, the empirical finding that corporate CEOs are taller than average and male, whether in Sweden or among the Fortune 500. Saying “we need more startups for more jobs” is tantamount to saying, “we need more tall, male CEOs for more Fortune 500 companies.”

There is also a different kind of causality issue: As I have argued elsewhere, all successful startup-rich regions had quite large corporations in their pasts infusing the ecosystem with talent, connections and knowledge: Boulder had IBM and nearby NORAD; Helsinki had Nokia; Israel had Tadiran and the Lavi project; Silicon Valley had Fairchild; Waterloo had Research in Motion; Bangalore had IBM; and Boston had Raytheon and MITRE. Later, each of these regions evolved a strong startup scene. But those few startups that grew arguably were results before they were causes.

The methodology masks mirages. Furthermore, some of the empirical support for the idea that “startups create jobs” results from a kind of statistical illusion, or what we call an artifact. Let me explain. Imagine you have two dice: the Startup Die has the numbers 0 through 5 on each of its sides. You roll the Startup Die and the number indicates the number of jobs you have created. If you roll the Startup Die a few hundred times, each roll a startup, then randomly the average number of jobs created will be a positive number. Voila, “startups create jobs.” But if you are an established company with many employees, then your Established Die includes negative numbers, let’s say odd numbers between 5 and -5 (i.e. -5, -3, -1, +1, +3, +5). Now if you roll the Established Die a few hundred times, you will get zero. Voila, big companies don’t create jobs.

The list goes on. The problem is not in the research or researchers – they are reporting empirical facts and often dealing with complex and incomplete data sets. The problem lies with those translating the research unthinkingly and uncritically into policy and practice. There is no doubt that some – a very small percentage – of startups, do indeed create some good jobs. But public and business leaders as well as policymakers in the U.S. and elsewhere must see startups accurately and in perspective in order to foster growth and long-term economic prosperity.