Common MLB Statistics: Which Stats Determine a Team’s Win Percentage?

Sarah Sult
12 min readMay 9, 2021


The following is a paper I wrote for my Introduction to Econometrics course in the Spring of 2021. The paper is a regression analysis of a topic of our choice. I wanted to examine which mainstream statistics are most indicative of a team’s success (measured by their win percentage). I had a wonderful time learning more about how economists examine their data and contrasting that with what I have learned in my statistics courses.

I. Introduction

At the basis of every sport is the desire to win. Athletes love to say they play for the love of the game, but the bottom line is they love to win the game.

Baseball leads most professional sports in its developed statistical measures of players, teams, and the mechanics of the game. In 1977 Bill James, the founder of sabermetrics, began writing about the underlying strategy of baseball and how more mathematically and analytically minded people could measure different aspects of the game. Members of the baseball community did not take well to James’s work at first. Many people adhere to the “don’t fix what isn’t broken” way of thinking.

It was not until the 2002 season, the Moneyball era, that people began to accept what analytics could bring to the sport. Athletics’ General Manager Billy Beane built his team based on a player’s ability to get on base rather than any other measure of his skill. Even throughout the 2002 season, Beane clashed with the strategies of more “old-school” scouts. The A’s went on to win their division with a 103–59 record. They even won 20 straight games in August and September. Beane was a fan of James’s work, and with the success of his 2002 team, was able to bring sabermetric analysis into many front offices throughout the league.

James created the Pythagorean Win Percentage to measure expected win percentage based on the ratio of a team’s wins and losses to runs scored and allowed. While the win percentage is an effective measure of a team’s winning percentage, it does not indicate what characteristics of a team lead to their success. James also created Game Score, a simple box-score-like measure of a starting pitcher’s performance any given day. Game Score gives a quick insight to fans as to where their team’s starting pitcher excelled and where they faltered.

I conducted an analysis on what team statistics contribute to their win percentage in a given year. The analysis will help give a bird’s eye view, similar to Game Score, about which statistics are most indicative of a team’s success. Variables include batting average, earned run average, runners left on base, and on base percentage among other more advanced statistics and numerical values representing the success of teams. I include both offensive and defensive statistics to determine the relative importance of both in determining winning percentage for a given season.

II. Prior Research and Theory

The majority of regression studies for baseball are based on the finances. For example, what determines what a player will be compensated or what a team will make in a season. Regression analyses on performative aspects of the game are less common.

Houser (2005) wrote about his regression analysis on which team statistic is most important in determining team success. His dependent variable was number of wins, along with the majority of other studies on team success. I chose win percentage because it is not affected by the number of games played (although all seasons my analysis consisted of the usual 162 games in a MLB regular season). Houser found the most important variables were on-base percentage (OBP), walks plus hits per inning pitched (WHIP), and slugging percentage (SLG), all of which I included in my data. OBP and WHIP are both measures of how players get on base. A high OBP and a low WHIP indicate a team is good at getting on base while keeping opposing teams off. Houser’s results are in line with Beane’s strategy for the 2002 Oakland A’s. Houser also concludes more generally both offense and defense contribute to wins.

Fullerton et al. (2014) conducted an econometric analysis of the 2013 MLB season, also with the dependent variable of regular season wins. Their results emphasized the importance of solid team pitching, defense, and offense. They included a financial analysis I did not want to delve into with my data.

Pavitt (2011) emphasized the division of defense and offense and found which, if any, was more important in determining wins. He defined defense as consisting of fielding and pitching, offense just included hitting and no baserunning statistics. When defined this way, Pavitt concluded defense and offense have equal effects on team success. He argued the equal effects were expected as baseball is a balance between runs a team scores and runs they prevent the other team from scoring.

Previous literature is consistent in stating offense and defense are both important in team performance. Since all team sports emphasize offensive and defensive production, equal importance makes sense.

III. Data

The data set I used was cross sectional. It included all 30 MLB teams over 4 different years. I chose the non-consecutive years 2011, 2014, 2016, and 2019 and took hitting and pitching statistics from Baseball Reference, which totaled 120 observations for each variable. I regressed the dependent variable WINP (win percentage) over 10 independent variables. All statistics are calculated using only regular season results. All statistics and values are cumulative for the whole team and the whole season. The variables and their expected signs can be found in Table 1.

Table 1 — Variable Descriptions and Expected Signs

Dependent Variable:

WINP — the total regular season wins for one season divided by the total number of regular season games multiplied by 100 to get a percentage.

Independent Variables:

BA — Team batting average calculated by (hits / by at bats). Teams with a higher batting average have more opportunities to score. I would expect BA to have a positive sign.

BB — Total walks by the hitters of each respective team. Teams whose players walk more get on base more and have more opportunities to score. I expect BB to be positive.

ERA — Earned run average, calculated by (9 x earned runs / innings pitched). Higher ERA’s for teams mean other teams are scoring on them more. I expect ERA to be negative.

FIP — Fielding independent pitching, calculated by ((13*HR)+(3*(BB+HBP))-(2*K))/IP + constant (constant brings onto ERA scale). FIP is similar to ERA. FIP measures the amount other teams are scoring on them but removes any effects of fielding. I would expect FIP to have a negative value as well.

IBB — The total number of intentional walks of a team’s batters. Opposing teams’ pitchers can elect to intentionally walk a batter so they do not have to face them at the plate. While walking a batter gives the hitting team more runners on base, it is usually used to face a less productive hitter to hopefully end the inning. The conflicting affects make me unsure of what sign IBB will have in the regressions.

LOB — The total number of runners left on base. The more runners a team leaves on base, the less they take advantage of their opportunities to score. I would expect LOB to have a negative value.

OBP — On base percentage, calculated by (Hits + Walks + Hit by Pitch) / (At Bats + Walks + Hit by Pitch + Sacrifice Flies). The more players get on base, the more chances they have to score (the entire basis of the Moneyball empire). I would expect OBP to have a positive value.

SB — The total number of bases stolen by a team’s hitters. Stolen bases give runners opportunities to get in better scoring positions. I would expect SB to have a positive sign.

SLG — Slugging, calculated by (1B) +( 2 x 2B) + ( 3 x 3B) + ( 4 x HR) / AB. Slugging represents the number of bases recorded per at bat. SLG differs from BA because it does not include walks or hit-by-pitch and does not value bases evenly. I would expect SLG to have a positive sign.

WHIP — Walks and hits per innings pitched, calculated by (Walks + Hits)/Innings Pitched. A higher WHIP means the opposing team has more opportunities to score. I would expect WHIP to have a negative value.

Table 2 lists the descriptive statistics for each variable. BA, OBP, and SLG are all percentages, and as such, are between 0 and 1. BB, IBB, LOB, and SB are all in total number of bases, and are much larger numbers than percentages and statistics scaled to innings.

Table 2 — Descriptive Statistics

IV. Regression Analysis

I ran the make some equations program on the linall equation in EViews. Linall is a least squares regression linear in all the variables. Out of all the regressions, I excluded the log equations from further consideration because the superset failed the RAMSEY test, and the linear regressions did not.

Table 3 — Best Regressions

I marked all the top three values for each selection criteria in the excel worksheet. A01SAF5 and A01SCF3 had top 3 selection criteria values, passed the RAMSEY, and had the superset pass the RAMSEY. A01SAF5 is a forward stepwise regression with a p-value of 0.5 of linear WINP on all the fractional polynomials of the RHS. A01SCF3 is the same but with a p-value of 0.3. Passing the RAMSEY means A01SAF5 and A01SCF3 have the correct specification. A01SAF5 had the lowest Akaike, the third highest adjusted R squared, and the third lowest Schwartz and HQ. A01SCF3 had the lowest HQ and the second lowest Schwartz. Both regressions also passed all heteroskedasticity tests, meaning all reported statistics are valid.

I conducted some coefficient analysis to determine which of the two regressions I wanted to use as my final regression. Every variable in A01SCF3 is statistically significantly different from zero, while not every variable is significant in A01SAF5. A WALD Coefficient Restriction Test on all BA variables in A01SAF5 shows the variables are jointly significant at the 1% level and on all FIP variables shows the variables are jointly significant at the 5% level. Since all of the variables are significant, even if the data has multicollinearity, there is no multicollinearity problem.

Graph 1 — A01SAF5 Univariate Part FIP

I viewed the univariate part of each variable in A01SAF5 first. The univariate part of FIP (see Graph 1) convinced me A01SAF5 was not an appropriate regression to represent effects on WINP. FIP is similar to ERA because lower scores represent better performances. I would expect FIP to be a linear graph with a negative slope, not what is pictured above. FIP was not included in A01SCF3, and the other univariate parts of A01SCF3 appeared as would be expected based on theory. I decided to stick with A01SCF3 for the full analysis.

A WALD test on all variables omitted from A01 to A01SCF3 show that the variables are jointly insignificant at the 10% level.

A01SCF3 includes polynomial fractions for ERA, LOB, OBP, SLG, and WHIP. I was surprised BA was not included in the final regression because it is such a common statistic, but as baseball analytics conversations are moving forward, OBP seems to be the more preferred statistic over BA.

Table 4 — ERA Slope

The univariate part of ERA is a monotonic downward sloping function with a slope at the mean of -8.26. For every singular increase in ERA, the team’s WINP is expected to decrease by 8.26%. Theory supports the decrease because as ERA increases, the opposing team is scoring more.

Table 5 — LOB Slope

I was pleased LOB remained in the final equation. LOB is not a value used frequently when discussing wins and win percentage though it plays a huge role in a team’s success. Every runner left on base is a missed opportunity to score. The univariate part of LOB is a monotonic downward sloping line with a slope at the mean of -0.018. For every singular increase in LOB, the team’s WINP decreases by 0.018%. While the effect seems small, the average total LOB for all 30 MLB teams was 1107.317. The effects add up over time.

Table 6 — OBP Slope
Graph 2 — A01SCF3 Univariate Part of OBP

The univariate part of OBP is a non-monotonic hat shaped graph. The maximum value for OBP is .352, and since the graph does not appear to begin to slop downward by an OBP of .352, I am going to treat it as a monotonic function. The slope at the mean is 178.1385, but the graph flattens out at about an OBP of 0.34. The slope is large because WINP is measured as a percentage and OBP is measured as a proportion. A one increase in OBP would increase WINP by 178.14%. Both values are off the scale of proportions and percentages. For reference, I included Graph 2, the graph of the univariate part of OBP. Regardless of specific values, the increasing effect of OBP is in line with theory because getting on base increases the chances of a team scoring. The tapering off at about 0.34 indicates past a certain point, increasing OBP would not have an effect on a team’s win percentage.

Table 7 — SLG Slope
Graph 3 — A01SCF3 Univariate Part of SLG

The univariate part of SLG is a monotonic upward sloping line with a slope at the mean of 102.848. The slope would indicate a one increase in SLG would lead to a 102.84% increase in WINP. Similar to OBP, the values are off the scales of proportions and percentages. I included Graph 3, the univariate part of SLG.

Table 8 — WHIP Slope
Graph 4 — A01SCF3 Univariate Part of WHIP

The univariate part of WHIP is just slightly not a monotonic function. The graph appears parabolic and increases slightly before hitting a maximum at about 1.15 and decreasing. Considering the minimum WHIP in the data set 1.102, I am going to treat the graph as a monotonic function and use the slope at the mean. I did include Graph 4 for reference. The slope at the mean is -17.25. For every singular increase in WHIP, the team’s WINP decreases by 17.25%. While the percentage change seems large, the results are on an entire 162 season span. A team with a 1.5 WHIP at the end of an entire season did decidedly better than a team with a 2.5 WHIP.

Regression A01SCF3 indicates the most important statistics of the ones included in the data are ERA, LOB, OBP, SLG, and WHIP. Houser’s (2005) work supports the significant variables as he found OBP, SLG, and WHIP to be the most important statistics in determining wins. Houser had an R squared of .849 for his regression. My regression had an R squared of .889, which is consistent with his findings. Similarly to Pavitt (2011) and Fullerton et al. (2014), my final regression also includes a balance of offensive and defensive statistics, emphasizing the importance of both in determining the success of a team.

V. Conclusions

The results from my final regression supported what I expected from theory and prior research. The final significant variables were ERA, LOB, OBP, SLG, and WHIP. The significant variables indicate the importance of both offensive and defensive strength for the success of a team. OBP, SLG, and WHIP were specifically mentioned in previous research, and ERA is a common pitching statistic.

LOB was the one variable in the final regression not common in prior research. I included LOB out of my own personal interest of its effects on WINP. The presence of LOB in the final regression means it is a value worth considering more often when discussing the success of a team. LOB is mentioned frequently during broadcasts but is not something analysts discuss much otherwise. Every runner left on base is a missed opportunity to score. I believe runners left of base could be an industry inefficiency worth exploring more, much like Beane did with on base percentage with the 2002 Oakland Athletics.

VI. References “2011 Major League Baseball Season Summary.” Accessed April 9, 2021. “2014 Major League Baseball Season Summary.” Accessed April 9, 2021. “2016 Major League Baseball Season Summary.” Accessed April 9, 2021. “2019 Major League Baseball Season Summary.” Accessed April 9, 2021.

Fullerton, Steven L, Thomas M Fullerton, and Adam G Walke. “An Econometric Analysis of the 2013 Major League Baseball Season,” n.d., 7 (2014).

Hallmark, Will. “Estimating Wins.”

Houser, Adam. “Which Baseball Statistic Is the Most Important When Determining Team Success?,” n.d., 10 (2005).

Passov, August. “What Statistical Factors Determine Success in the MLB?”

Pavitt, Charles. “An Estimate of How Hitting, Pitching, Fielding, and Basestealing Impact Team Winning Percentages in Baseball.” Journal of Quantitative Analysis in Sports 7, no. 4 (January 27, 2011).



Sarah Sult

WUSTL Economics and Computer Science major | baseball analytics enthusiast