Secrets of Game Engagement

Engagement Insights from Steam Data

In order to uncover the hidden systems behind game engagement, we analysed a large Steam dataset containing a wide range of usage metrics.

Analysing the Steam Dataset Engagement Does Not Mean Scale Using a Median Average The Outperforming Games

What makes games engaging? Why do some titles retain players for hundreds or even thousands of hours, while others are quickly abandoned? In this article, we begin our series exploring the Secrets of Engagement. 

To examine engagement more closely, we turned to a large public dataset of games on Steam, compiled via Hugging Face. The dataset aggregates information on roughly 150,000 titles, including estimated ownership ranges, player counts, and average lifetime playtime.

While no dataset of this kind is perfect, it provides a useful large-scale view of how games are actually played in the wild.

Preparing the Data

Before analysing the dataset, we applied a small number of straightforward filters to ensure the results reflected typical player behaviour rather than technical artefacts or edge cases.

We excluded entries that were not conventional games, such as utilities, creative tools, and software packages, where recorded playtime may reflect background usage rather than active engagement. We also removed incomplete records and restricted the dataset to titles with valid ownership and playtime estimates.

To reduce the influence of extreme values, we capped playtime at the 99th percentile. This did not materially alter the overall structure of the dataset, but prevented a very small number of outliers from distorting the wider distribution. We also restricted the analysis to English-language titles only.

How to Measure Player Engagement?

At first glance, answering this question is fairly straightforward. However, establishing a reliable metric is more complicated than it first appears. Depending on the approach we choose, very different kinds of games rise to the top:

For these reasons, playtime per player is the most useful metric, and this will form the basis of our analysis.

Ownership and Player Engagement 

The Steam dataset used for our analysis includes a metric called ‘average lifetime playtime’, which refers to the average amount of time players spend in a game over the period they own it. Conceptually, this functions as an average playtime-per-player metric, allowing comparison across games of different scales.

Our first graph (Chart 1) plots average lifetime playtime across the different ownership tiers already present in the Steam dataset, based on estimated player ownership. Each point represents a game.

Several patterns are immediately apparent. Most importantly, engagement varies enormously within every ownership tier. Games with broadly similar player-base sizes can exhibit dramatically different levels of engagement.

Some smaller titles have extremely high levels of engagement, while certain larger releases attract comparatively modest attention. This suggests that player-base size alone cannot explain why some games retain players far longer than others.

Towards a Comparative Cohort Analysis

Although this playtime-per-player metric provides a useful indicator of engagement, ranking games purely by this measure would still produce a relatively narrow view of the market. The analysis would likely become weighted towards a small subset of games and audience behaviours.

Instead, we wanted an approach that could surface engaging games across different levels of popularity and commercial scale, providing a broader view of engagement across the market.

To do this, we used the ownership tiers as comparative cohorts. These tiers group games according to estimated player ownership, allowing us to compare titles operating at broadly similar levels of commercial reach.

A Median Baseline for Each Cohort

As shown in Chart 1, the Steam dataset segments games into several ownership tiers, ranging from titles with fewer than 100k owners to games with more than 10 million.

For each ownership tier, we calculated the median average lifetime playtime across all games within that cohort - in other words, the midpoint engagement level for games with similar levels of player ownership.

This established a baseline level of ‘typical’ engagement for each cohort, allowing us to identify titles that retain player attention significantly more effectively than their peers. Chart 2 plots these median engagement levels across the different ownership cohorts, providing a baseline against which individual games can be compared.

Deriving an Outperformance Ratio

Having established a baseline level of typical engagement for each ownership cohort, we wanted to identify games that significantly exceeded this benchmark, in other words, titles that ‘overperform’ relative to other games operating at a similar level of ownership.

To do this, we derived an outperformance ratio by dividing a game’s average lifetime playtime by the median average lifetime playtime of its ownership cohort:

Outperformance ratio = game average lifetime playtime ÷ cohort median average lifetime playtime

A value above 1 indicates that a game sustains higher-than-typical engagement relative to other titles within its ownership tier.

When these outliers are plotted against the wider dataset, a clear pattern emerges. Across every ownership tier, from smaller independent releases to large-scale commercial games, there are titles that have unusually high levels of engagement relative to their peers.

These are not simply the biggest games, nor the most commercially visible. They are the titles that, for whatever reason, succeed in holding player attention far beyond the norm for their cohort.

We plotted these relationships on a third graph (Chart 3), comparing individual games against the median engagement baseline for their ownership tier.

The line represents the median level of engagement within each ownership tier, providing the baseline against which individual games can be compared. Games that sit close to this line perform broadly in line with the typical engagement level for their cohort, while those below it attract comparatively lower levels of sustained player attention.

The most interesting cases, however, are the titles that sit significantly above the line. These games, highlighted as purple dots, have unusually high levels of average lifetime playtime relative to other titles operating at a similar scale of ownership.

To make these patterns easier to interpret, the lowest ownership tier (<100k owners) has been excluded from this chart. Because the median engagement level within this category is extremely low, a very large number of titles appear to overperform relative to the cohort baseline, reducing the usefulness of the comparison. Even with this category removed, a clear pattern remains. Across every ownership tier, from smaller independent releases to large-scale commercial games, there are titles that consistently retain player attention far more effectively than their peers.

A Set of Outperforming Games

From the dataset, we compiled a table listing the strongest-performing titles within each ownership tier: games that sustain unusually high levels of player engagement relative to other releases operating at a similar scale.

The key figure is the right-hand column: the ‘outperformance ratio’, which expresses how much a game’s average lifetime playtime exceeds the median playtime for its ownership cohort.

As the table shows, the smallest games in our analysis (100k-200k owners) tend to exhibit the highest relative outperformance. This pattern continues across the wider dataset, with the ratio gradually declining as ownership tiers increase in size.

One possible explanation is structural. Steam contains a very large number of smaller titles, many of which attract little sustained engagement. This pushes the median playtime of lower ownership tiers downward, making exceptional performers within those categories appear more dramatically above baseline.

Clues to Player Engagement

These overperforming titles, games that significantly exceed the typical engagement level for their ownership cohort, offer useful clues as to the drivers of sustained player attention.

Rather than simply reflecting commercial scale or visibility, they point towards underlying design characteristics that repeatedly encourage players to return over time.

In our next article, we examine fourteen of these games in greater detail, adding a qualitative layer based on player comments and professional reviews to better understand what makes them so engaging. 

Series: Secrets of Engagement

The Hidden Systems of Engagement Engagement Insights from Steam Data Five Systems of Engagement Series Conclusion: What Keeps Players Playing? 1. Mastery Engines 2. Generative Social Systems 3. Creative Stewardship Systems 4. Persistent Stakes Systems 5. Fantasy Anchors