The (slightly filtered) dataset underlying this post has been released publicly on PsyArxiv. If you want me to do more analyses, please contact me any time. You can do them yourself, too, of course!

This is part 2. Re-read part 1 here.

I finally took the time to get into the data and do the first analyses for which I used the free R, a free software environment for statistical computing and graphics. First I removed outliers – people who indicated impossible hours played per week (cutoff at 112h) and missing data were the most commen exclusion criteria. This reduced my sample from 14250 to 14067 participants.

Who took the survey – some descriptives

From these I calculated the most played games. If a participant indicated more than one game I only evaluated the first mentioned game (because I asked to only specify one game).

The ten most played games were:

• League of Legends (11564)
• Starcraft 2 (337)
• Counter Strike (280)
• World of Warcraft (142)
• Hearthstone (99)
• Diablo 3  (84)
• Dota 2 (42)
• Heroes of the Storm (39)
• Guild Wars (37)
• Skyrim (36)

I grouped the rest into Other (405). Note that in the survey the “Game” filed was a free-text field, so everybody could spell their game however they liked. I only evaluated spellings of games which were given by at least two people to reduce workload (left me with 13055 participants). Also note that this uneven distribution of participants across games might make cross-game analyses difficult (we’ll come to this later).

Is more playing predictive of less happiness? Global correlation analyses

Remember the three scales I employed. The Satisafaction With Life scale and Generalized Anxiety Disorder 7. A correlational analysis reveals how much the values in two or more data sets are varying together. The question is, to which extent can the variations in one dataset be explained by the variations in the other dataset?

I found out that the more hours you play, the less satisfied with your life you tend to be (r = -.13, p < .001)If you don’t understand the numbers, trust my word or look at What do the numbers mean? at the end of the article. The relationship of “hours played” and social phobia is as I expected. People who play a lot of games are likely to be socially more isolated than people who don’t. This is not exclusive to gaming but any hobby one pursues on their own taken to an extreme (like binge watching TV). More interestingly,  a lot of time spent on gaming may be due to unhappiness in life in general. With such correlational analyses alone we are unable to make any causal conclusions at all. This means that I cannot say whether people play more because they are unhappy or are unhappy because they play more. To understand how correlation does not imply causation, have a look at these things that correlate although they have nothing to do with each other. Both might be the case, although I am inclined to believe that unsatisfactory life conditions might cause people to isolate themselves more and play more video games. To visualize correlations with many data points is unfortunately not easy. I tried it with hexbins. What you can see in the graph below is the number of people for various hours played (x-axis) plotted against their perceived Satisfaction With Life (y-axis). The participants are pooled into hexagons, the color of which gives information about how many people fall into this specifc hexagon.

Thankfully, a user on reddit made me aware that there is a better way to look at the data. I divided up participants according to the hours played scale into 10 equal groups of 11.2 hours. So the first group played between 0 and 11.2 hours, the second group played between 11.2 and 22.4 hours and so on. For each of these groups I calculated the average SWL score. What you can see in the graph below is, that there is a consistent downwards trend (see the smoothed trendline in red) across the whole sample. The more you play, the less satisfied you tend to be. Except for people who played more than 101 hours per week apparantly but I only had a few participants in this group and they are likely not quite representative of their group.

Now also interesting is the last scale: Generalized Anxiety Disorder 7. Generally by filling it out, you can score between 0 points (no GAD) and 21 points (severe GAD). Using the cutoff score of 10, the GAD‐7 has a sensitivity of 89% (11% of GAD patients are not diagnosed) and a specificity of 82% (18% of healthy participants are falsly diagnosed as having GAD) of detecting generalized anxiety disorder. This means it’s fairly good at detecting GAD. I grouped my participants according to the cutoff score 10 in “GAD” and “no GAD” and found that people with GAD play 23.9 hours per week on average while people witout GAD play 20.7 hours per week. I find it a little bit concerning that over 20% of the people who filled out the survey fell into the GAD group. This is way more than I expected. But anxiety currently afflicts more than 20 million Americans, making it the most common mental illness in the US. If you are worried you might suffer from it, start with an internet self test.

On a lighter note, I also checked gamers of which game are the happiest and/or play the most. Note that the Satisfaction With Life scores only differ by about 10% in total, so the differences are rather small.

In the next blog post I will look into gender, work status and narcissism. A statistical comparison of games is almost impossible because of the participants being almost exclusively from the LoL community. I might have enough Starcraft 2 players for a comparison with League of Legends. I will see what I can do about that. Questions and Comments appreciated!

What do the numbers mean?

Example: “the more hours you play, the less satisfied with your life you tend to be (r = -.13, p < .001)”

The p-value is related to the concept of statistical significance which is a mathematical technique to measure the strength of evidence from a single study. Statistical significance is conventionally declared when the p-value is less than 0.05. The p-value is the probability of seeing a result as strong as observed or greater, under the null hypothesis (which is commonly the hypothesis that there is no effect). Thus, the smaller the p-value, the less consistent are the data with the null hypothesis under this measure. (see here)

The r-value is a correlation coefficient, an estimate of how big the relationship between two sets of data is. In the example above, we look at the relationship between “hours played” and “satisfaction with life.” With a correlational analysis we look at how much of the fluctuations in the one data set are similar to fluctuations in the other. The lowest r-value is 0 correspong to no correlation, a total correlation is 1 (or -1). So ist a r-value of .13 describing a meaningful relationship? To answer this we need to think about whether “hours” played” is the only thing that can influence life satisfaction. Definitely not. There are dozens of things which could influence life satisfaction, for example income, relationship status or health. So a 13% relationship seems pretty meaningful.