What determines the price of games? #2
Once again, I welcome you to a brief investigation about games, gamers, and the gaming industry. Today, we will look at the factors that affect people's willingness to pay for games. The overall theoretical framework I'm trying to confirm is layed out here.
My previous article on the matter dealt with active players as a possible variable affecting a game's price. A look at 15 games suggested that the more people actively play a game, the higher its price is likely to be. This time I will consider a variety of other variables that might contribute to game prices. Don't worry now - you won't be asked to understand any of the statistical mumbo-jumbo used to analyze SteamSpy data - the core of this article will be about the theories I've formed from my investigation.
Getting my hands on the numbers
To figure out what factors might affect the price of Steam games, I went into Steamspy and collected data on 52 games in total, 36 of which are not free-to-play games. I put the raw data here, if you want to have a look. The 13 variables describing each game range from genre and active player count all the way to 3D/2D, full release/beta, and user verdicts.
An overall review
For the first viewing, I made a series of simple regressions in Excel, much like in the previous article. From this, it was clear that some of the variables have notable influence in terms of price determination. But some definitely do not.
Here are some graphs to skim over, joined with quick verdicts as to whether or not the variable is a plausible influence on price or not.
As you can see, insignificant variables include competition and overall consumer spending in the games genre - and strangely, developer experience. One interesting variable here is the user review score - it turns out that, the higher the game's price, the lower the user review is likely to be. This tells us that game reviewers are very keen at identifying value for money - good news for free-to-play developers!
The most significant variable from this first observation is the amount of owners playing regularly. This is a constructed metric, but it clearly shows that the more regularly owners of a game play it, the more likely the price of the game will be high. Excellent deduction - the more enjoyed the game is, the higher the price. This probably has less influence on initial pricing that it does on price drops after release. Let's move on.
Sketching a proper model
With this in mind, I set out to make a multi-regression model for the data, essentially combining several of the little models together to see if the fit improves. There were 13 variables initially, and I calculated 2 more - active players over copies sold and years since release. Unfortunately, only a few of these proved statistically significant. For the final model, I left only the variables that were either significant or surprisingly insignificant - the rest are just generally not noteworthy. You might or might not be familiar with SPSS, so I will explain the output table below.
What you can see in the 5 columns is as follows:
Firstly, there are the variables used, with the constant (not important) on top. The second column shows the coefficients for each variable. For example, if a game is a beta, it is estimated to be $13 less expensive than an equivalent full release. The third column tells us the standard error - how much do these estimates vary over the sample? The fourth is the t-statistic, which is just a ratio between the coefficient and its standard error. The t-statistic then lets us infer the significance or p-value - if the value in the fifth column is below 0.05, we assume the variable is indeed statistically significant in forming the price.
How the p-value works:
The p-value measures "the amount of evidence in favor of the assumption that the coefficient is zero". In other words, the lower the p-value, the less likely it is that the coefficient is zero - the more reason to doubt that the variable is absolutely insignificant. Usually, we assume that a p-value of below 0.05 (or 5%) is significant enough for us to say that the estimated coefficient indeed tells us something about the population.
What does the table tell us?
1. As mentioned before, if a game is a beta, it is expected to be around $13 cheaper than a similar full release.
2. On top of that, for each percent of owners who actively play the game, the price increases by $3.3 - the more people enjoy their purchase, the more new players are willing to pay to join in on the fun.
3. Here is a strange one that I mentioned before. The more expensive the game, the worse its review is likely to be, at around 1% for each $1. Note that this means some people are willing to pay high prices for games with bad reviews...
Taking a closer look reveals that there is little connection between user reviews and copies sold or active players, implying that consumers don't give as much weight to the opinions others as one might think.
4. Having a solid single-player mode available turns out to increase a games price by $12.5, while having a multiplayer mode actually has no significant effect.
5. Speaking of insignificant effects, it does not seem to matter how old the game is, how experienced its developers are, or how many spatial dimensions it offers for exploration.
There is plenty more to be done with this sample, and I did not even start working with the 16 free-to-play games. However, we can already find a little something for conclusions.
First of all, Steam user behavior is quite bewildering - some of the variables that you'd think would matter seem completely insignificant, such as amount of in-game dimensions and developer experience, and things like user scores display tremendous negative causation. It can be universally recognized, then, that reviewer and consumer verdict is not that important to Steam users. On top of that, users expect to pay much less for betas and games without a proper single player experience.
I hope this gives you some food for thought; it certainly gives me some. Untangling the mystery of game pricing is turning out to be much harder than expected. Future articles will certainly include an investigation into free-to-play games, and may add average playtimes and re-playability into the mix of variables.
Hope you enjoyed this article! Let me know your thoughts in the comments below.