When I started my Master's degree in January 2018, I was confident I would be done in a year and half. After all, I only had one year of classes and I figured 6 months to write a thesis would be plenty.
Three years later, I'm finally done: the final version of my thesis was accepted on January 22nd 2021.
My thesis, entitled What are the incentive structures of Free Software? An economic analysis of Free Software's specific development model, can be found here 1. If you care about such things, both the data and the final document can be built from source with the code in this git repository.
Results and analysis
My thesis is divided in four main sections:
- an introduction to FOSS
- a chapter discussing the incentive structures of Free Software (and arguing the so called “Tragedy of the Commons” isn't inevitable)
- a chapter trying to use empirical data to validate the theories presented in the previous chapter
- an annex on the various FOSS business models
If you're reading this blog post, chances are you'll find both section 1 and 4 a tad boring, as you might already be familiar with these concepts.
So, why do people contribute to Free Software? Unsurprisingly, it's complicated. Many economists have studied this topic, but for some reason, most research happened in the early 2000s.
Although papers don't all agree with each other and most importantly, about the variables' importance, the main incentives2 can be summarized by:
- expectation of monetary gain
- writing FOSS as a hobby (that includes “scratching your own itch”)
- liking the FOSS community and feeling a sense of belonging
- altruism (writing FOSS for Good™)
Giving weights to these variables is not an easy thing: the FOSS ecosystem is highly heterogeneous and thus, people tend to write FOSS for different reasons. Moreover, incentives tend to shift with time as the ecosystem does. People writing Free Software in the 1990s probably did it for different reasons than people in 2021.
These four variables can also be divided in two general categories: extrinsic and intrinsic incentives. Monetary gain expectancy is an extrinsic incentive (its value is delayed and mediated), whereas the three other ones are intrinsic (they have an immediate value by themselves).
Theory is nice, but it's even better when you can back it up with data. Sadly, most of the papers on the economic incentives of FOSS are either purely theoretical, or use sample sizes so small they could as well be.
Using the data from the StackOverflow 2018 survey, I thus tried to see if I could somehow confirm my previous assumptions.
With 129 questions and more than 100 000 respondents (which after statistical processing yields between 28 000 and 39 000 observations per variable of interest), the StackOverflow 2018 survey is a very large dataset compared to what economists are used to work with.
Sadly, it wasn't entirely enough to come up with hard answers. There is a strong and significant correlation between writing Free Software and having a higher salary, but endogeneity problems3 made it hard to give a reliable estimate of how much money this would represent. Same goes for writing code has a hobby: it seems there is a strong and significant correlation, but the exact numbers I came up with cannot really be trusted.
The results on community as an incentive to writing FOSS were the ones that surprised me the most. Although I expected the relation to be quite strong, the coefficients predicted were in fact quite small. I theorise this is partly due to only 8% of the respondents declaring they didn't feel like they belonged in the IT community. With such a high level of adherence, the margin for improvement has to be smaller.
As for altruism, I wasn't able get any meaningful results. In my opinion this is mostly due to the fact there was no explicit survey question on this topic and I tried to make up for it by cobbling data together.
Kinda anti-climatic, isn't it? I would've loved to come up with decisive conclusions on this topic, but if there's one thing I learned while writing this thesis, it is I don't know much after all.
Note that the thesis is written in French. ↩
Of course, life is complex and so are people's motivations. One could come up with dozen more reasons why people contribute to Free Software. The "fun" of theoretical modelisation is trying to make complex things somewhat simpler. ↩
I'll spare you the details, but this means there is no way to know if this correlation is the result of a causal link between the two variables. There are ways to deal with this problem (using an instrumental variables model is a very popular one), but again, the survey didn't provide the proper instruments to do so. For example, it could very well be the correlation is due to omitted variables. If you are interested in this topic (and can read French), I talk about this issue in section 3.2.8. ↩