Why Stats Matter In Evaluating A Coaching Career
You look at a coach and you see wins and losses, headlines, and drama. That is surface level stuff. The actual numbers tell a different story that most people skip over because they are boring or inconvenient. I have spent years building spreadsheets and analyzing coaching data, so I know what the raw numbers actually show when you strip away the narrative. Jim Harbaugh has coached at multiple levels, and each stop tells its own story. At Stanford from 2007 to 2010, he posted a record of 33 wins and 15 losses. That was a significant jump from the preceding years. He took a program that had not seen a winning season in a while and put it on a national map. The Sagarin ratings and SPI numbers from that era show a team that outperformed its recruiting class consistently, which is the kind of detail that does not make highlight reels. Then came the San Francisco 49ers from 2011 to 2014. His NFL record was 28 wins and 20 losses. He made the playoffs every year, including a trip to the NFC Championship game in his first season. The advanced metrics like Expected Points Added per play and turnover margin tell you something important. The defense was elite in his early years, and the offensive efficiency numbers were solid but not historically great. The narrative always focuses on the playoff exits, but the regular season performance was arguably one of the best in franchise history during that span.
His return to college football at Michigan starting in 2023 has been even more scrutinized. The win total and national attention are obvious, but the underlying metrics are where the real story lives. Let me walk you through how I actually approached this analysis, because the methodology matters more than the final numbers. I started by pulling play-by-play data from multiple sources. Then I cleaned it, standardized the definitions, and ran efficiency metrics across all three tenures. The problem most people have is they take a single season or a single metric and declare it definitive. That is lazy work. You need to look at trends, context, and adjusted numbers.
How To Build A Proper Coaching Statistics Analysis
First, you need reliable play-by-play data. I use a combination of public datasets and the NCAA official stats. For NFL games, I pull from Pro Football Reference and Next Gen Stats where available. The raw data is messy. You will find missing plays, mislabeled formations, and games where the play type is ambiguous. I spent about three weeks just cleaning data for a single coach when I first started doing this seriously. The next step is calculating adjusted efficiency. Raw yards per play means almost nothing on its own. You need to adjust for opponent strength, game context, and pace. I use a simple adjustment factor based on the strength of schedule from FBS and NFL sources. It is not perfect, but it removes a lot of the noise that makes casual analysis misleading. Here is a specific edge case I ran into that most people miss. When analyzing Harbaugh at Stanford, I noticed that the offensive production numbers spiked in 2010, but the defensive numbers actually declined slightly. If you only look at the overall win column, you might think the team was dominant on both sides. The underlying data told a different story. The team was winning close games because the offense found a way to score in critical moments, not because they were controlling games throughout. I had to dig into situational stats, third down conversions in the red zone, and clock management numbers to confirm what was actually happening. That insight changed my entire assessment of that season.
Get the Full Details
For the 49ers years, I tracked turnover margin heavily. The team won several games by flipping the ball and creating positive turnover differential. In 2012, the turnover margin was plus nine, which correlated directly with a 12 win season and a playoff run. Years later, when the turnover luck regressed, the record dropped accordingly. That is the kind of detail that explains more than the headline win total ever would. Metric selection is critical. I recommend focusing on these core numbers: adjusted net yards per play on both sides of the ball, third down conversion rate, red zone efficiency, turnover margin, and strength of schedule adjusted win expectancy. These give you a complete picture without getting lost in irrelevant stats.
Key Findings From The Data
Across all three tenures, Harbaugh teams consistently outperformed their recruiting rankings. This is measurable. At Stanford, the 2010 team was rated as a three star recruit heavy roster but played at an elite level. At Michigan, the 2023 and 2024 classes were highly touted, but the 2022 class that arrived in his first year was less decorated than expected and still produced a national championship run. The recruitment-to-performance gap is real and documented. Defensive performance varies more than people assume. The 49ers defense in 2011 and 2012 was historically good, ranking in the top five in points allowed and pressure rate. By 2014, those numbers had dropped significantly, and the team became more one dimensional on offense. The data supports the feeling that the later years felt different, and it was not just narrative. At Michigan, the offensive efficiency numbers in 2023 were among the best in program history. The rushing attack averaged over five yards per carry, and the passing game was efficient but conservative. The decision making in close games, particularly in the Big Ten Championship and playoff runs, is where the data gets interesting. Some choices look aggressive on the surface, but the expected point models suggest they were mathematically sound decisions given the game context.
One counter intuitive finding: Harbaugh teams tend to have higher variance in scoring. They can put up 50 points or struggle to score 20. The standard deviation of point output is wider than average for coaches with similar winning records. This means his teams are less predictable, which makes season long evaluation harder for casual observers. A single bad game does not mean the system is broken, and a single great game does not mean everything is working perfectly.
Limitations Of This Kind Of Analysis
Statistical analysis of coaching has real limits. Play-by-play data does not capture locker room dynamics, player development, or in game adjustments that happen outside the box score. You can measure third down efficiency, but you cannot measure whether a coach motivated a player to perform above their talent level. That part remains qualitative. Sample size is another issue. Harbaugh has coached at the NFL level for four seasons and at the college level for roughly seven seasons across two stops. That is enough data to see trends, but not enough to draw absolute conclusions about causality. Correlation does not equal causation, and some of the success may be tied to specific players or institutional support rather than coaching alone. The strongest recommendation I can make is to combine advanced metrics with film study. The numbers will tell you what happened, but they will not always tell you why. For a complete picture, you need both. If you only have one, your conclusions will be incomplete.
For anyone wanting to replicate this analysis, I used a combination of Python scripts for data cleaning, R for statistical modeling, and a lot of manual verification. The whole process for a single coaching tenure takes roughly 40 to 60 hours depending on data quality. Budget accordingly if you plan to do this properly.
