Welcome to the inaugural post from the Civ Science project! Here we will dive into the dataset (see the about page for more information) for the first time and see what we can learn about unlocking social policies in Civ V. The data we are using includes the number of social policies unlocked in 180 multiplayer Civ V games played by FilthyRobot. This number includes ideological tenets.
Multiplayer games can end at different times. Sometimes they go all the way to spaceships, whereas other games end in a Medieval bloodbath. For this reason looking at the final number of social policies unlocked is not fair. Instead we shall examine the number of social policies unlocked per era. This is a way of normalizing the rate of unlocking policies over the length of the game.
DISCLAIMER: This is my first dive into the data, so most of what I show here is not complicated. In fact, we will be looking almost entirely at correlations, and as all good skeptics should know, correlation does not mean causation! So bear that in mind, that this post is showing associations between things, not necessarily mechanisms.
Wonders, City States, Religion… and Poland!
First let’s look at which civilizations are associated with an increased number of social policies. Before we look at the data, are there any predictions we can make? The most obvious is that we expect Poland to come out on top. The Polish civilization has an ability that gives a free social policy each time you advance to the next era. This guarantees 7 extra social policies for a full-length game! We might also expect civilizations with bonuses to culture to obtain more social policies than others. Civilizations like Polynesia or Brazil have tile improvements that give extra culture for example, whereas Siam has a unique building that provides extra culture, as well as an ability that provides additional culture from city states. But predictions are ultimately futile, and data is real, so let’s take a look.
DARKEST GREEN = 5 or more games, MIDDLE GREEN = 3-5 games, LIGHTEST GREEN = 1-2 games
The above graph shows the number of policies obtained per era, for each civilization in the game. The civs are ordered by their performance in this test, from highest to lowest. The bars are shaded according to how many games are in the sample size. This means that the darker the bar, the more reliable the estimate. The horizontal line represents the median policies per era game (i.e. a measure of the average). The grey bars represent the standard error, which is basically a way of showing the variation between the games for each civ.
There’s a lot to look at in the above graph so let’s list some key observations:
- As we predicted, Poland does very well, scoring well above the median.
- The Civilizations at the bottom of the list are civilizations with no bonuses to culture.
- Civilizations that get a strong religion perform well, such as Spain, the Celts, The Mayans and Ethiopia.
- Civilizations that have bonuses to city state alliances (Siam & Sweden) also perform well.
- Egypt comes out top, performing better than Poland!
What does this all mean? One observation is that while Poland is obviously associated with more social policies, a strong culture game can match, or even surpass Poland’s UA (unique ability). Looking at Egypt, it’s tempting to suggest that building lots of wonders (which Egypt’s UA encourages) is the most potent source of culture in the game. It’s also no surprise to see Siam performing well, as Siam can receive huge culture yields from city state alliances. Religion is also a great source of culture, with buildings like Mosques and Pagodas providing added culture per city. It’s worth noting that in two out of the three Spain games in our dataset, FilthyRobot had a faith-yielding natural wonder, so these games would have had an extremely strong religion.
One of my predictions isn’t reflected in the data at all. I anticipated that civilizations like Brazil and Polynesia, which have tile improvements that yield culture would have more social policies. It turns out that these two civilizations perform similarly to average. Perhaps the bonus from these improvements is too mediocre, or too situational to really change the number of policies obtained in the game?
Another fun observation is that Venice does well (although there is only one Venice game in our dataset), which also makes sense! Founding cities increases the culture cost of new cities, and Venice cannot found cities! It therefore makes sense that the cheaper culture cost of policies might translate to an advantage in how many policies Venice can unlock.
Note that there are probably some red herrings in here, Portugal has no culture bonuses at all, and yet is high on the list. This is probably down to random chance, that FilthyRobot focused on his Culture output in those games, without having anything to do with the civilization’s bonuses in particular. This can happen in datasets, and it’s not really a surprise.
But what do these numbers ultimately mean? Can we translate “policies per era” into the number of policies enacted in a real game? To work this out, let’s assume that the civ at the bottom of the list represents the bare minimum number of policies. For the sake of argument this value is the consequence of playing a game with no focus on culture whatsoever, and no bonuses to help culture output. We then calculate how each civ deviates from this minimum, and multiply by 8 (the number of eras) to predict how many extra policies each civ would get from a full-length game.
This graph looks very similar to the one before, but actually serves as a really useful sanity check! As you can see, Poland gets 7.5 more policies than minimum, and this is exactly what we’d expect! Their UA guarantees 7 for a full-length game, which suggests our analysis is probably reflecting something close to the truth.
The important thing to note about Poland is the tiny spread of the error bars. While there is more variation in the number of policies obtained by Spain or Egypt, Poland is completely consistent. When you play as Poland, you get those policies guaranteed, which is of course why Poland is considered one of the best civs in the game!
Which policies are the most culture efficient?
Social policies cost culture to unlock, it’s a basic mechanic of the game. However, several policies provide culture, such as the Legalism policy in the Tradition tree, which provides a free culture building in 4 cities. There is also the Representation policy in Liberty, which reduces the culture cost of future policies. I wanted to examine which social policy trees were associated with an increased number of policies unlocked throughout the game. That is to say, which trees are the best at paying for themselves by providing enough culture to unlock further policies.
To address this, let’s split the policies into two broad groups: opening policies and “filler” policies. In competitive multiplayer (based on the unmodded civ V experience), the opening policy options are Tradition and Liberty. Honor and Piety are available to players, but almost never chosen because they don’t provide good enough bonuses to get ahead in the game. Rationalism is such a strong policy tree that players unlock policies in that tree almost every game. For that reason it’s not considered here. “Filler policies” are therefore the social policies that are unlocked after completing Tradition or Liberty, but before opening Rationalism (which is essential to remain competitive). Although available from the start, Honor and Piety are considered filler policies, because they are only ever taken after Tradition or Liberty in competitive multiplayer.
The horizontal line equals the mean number of policies per era across all games
The first thing to note is that Tradition and Liberty come out as roughly equal when it comes to the number of policies each seems to enable in a full-length game. This surprised me a bit, as I had a hunch that the culture and wonder-building bonuses of Tradition would outperform Liberty, but it appears I was wrong!
It’s not a surprise that Aesthetics comes out ahead in the filler policy group, providing roughly one extra policy per game. This makes sense as Aesthetics is the primary culture-boosting tree. Remember that correlation is not causation though! This data might also be telling us that FilthyRobot takes policies in Aesthetics when he has a strong culture game (and uses Aesthetics to maximize that advantage), and so this policy tree might not be causing that policy advantage.
Some policy trees appear to be associated with fewer policies, notably Honor and Exploration. Both of these trees are quite culture-poor, not providing any culture bonuses to pay back their cost. The Honor tree is also the tree that FilthyRobot likely opens when he is fighting a war. During war, all his production will be focuses on military for defense and so building guilds, wonders and culture buildings will be extremely low priority. This might also explain why Honor is associated with more culture-poor games.
I then looked for all possible interactions, that is to say combinations of policy trees that particularly stand out as yielding fewer or more policies. Only one is worth reporting here, which is the big culture boost that seems to be unlocked by combining liberty and piety. Opening Tradition and Piety is associated with no more social policies than average. On the other hand Liberty + piety gives a big increase in number of policies per era, equivalent to 2-3 extra policies per game!
So, if you are worried completing liberty and piety might prevent you from gaining other policies later in the game, perhaps the trade-off isn’t as bad as you thought! Another explanation however is that it’s not the piety tree itself that is boosting culture, but the strong religion (as we see with the high number of policies unlocked in Spain and Celts games). The liberty + piety combination might just be an indicator of when FilthyRobot is employing this wide religious strategy.
To examine the effect of founding a religion, I looked at the games where FilthyRobot founded versus didn’t found a religion. Games with a founded religion did provide a small increase in social policies, equivalent to one extra policy a game. However, it’s worth noting that if we randomly generated datasets of the same size, we would expect to see this difference purely by chance 8% of the time (the p-value), so this might not be meaningful. Besides, founding a religion doesn’t mean you don’t get any religious bonuses, as you can still benefit from religions founded by other players in the game. What the data really suggest, is that strong religions, with high faith output are likely to provide a culture boost. In fact, Liberty + Piety strategies are predicted to gain as many social policies as the Polish UA! This is likely through the purchasing of multiple culture-producing buildings like Mosques and Pagodas, or perhaps even culture from Holy Sites after completing the Piety tree.
Does any of this matter?
This analysis of social policy numbers has been fun, but one key question hasn’t been looked at at all. Does unlocking more social policies increase the likelihood of winning? After all, winning is what it’s all about! It’s hard to test this directly without separating policies from all the other advantages in the game. For example, building wonders will boost culture, but also provide many other advantages towards winning. However, we can look at how many social policies were enacted in the games FilthyRobot won, versus the ones he lost:
Games won: 2.50 social policies per era
Games lost: 2.28 social policies per era
This difference might not sound like much at all, insignificant even, but bear in mind that this difference approximates to nearly 2 extra social policies per game. Think about how some of the strongest social policies are the hardest to unlock at the end of policy trees (Purchasing great scientists with faith, or purchasing spaceship parts with gold for example), and the magnitude of this difference becomes more obvious.
It is also worth noting that we only expect to see such a big difference purely by chance in 0.03% of datasets of the same size, which strongly suggests this difference is not down to luck.
Congratulations to making it to the end! Please do leave feedback, comments, questions, and suggestions for other things you’d like to see addressed with real data and statistics!