Wins Above Replacement and the MLB MVP Vote: A Natural Experiment Shane Sanders∗1, Joel Pottery2, Justin Ehrlichz1, and Justin Perline1 1Syracuse University, Sport Analytics 2University of North Georgia, Economics October 15, 2019 1 Introduction Major League Baseball was formed as a confederacy of two leaguesthe National League (NL; 1876- ) and the Amer- ican League (AL; 1901- )in 1903. Since 1911, the NL and AL have chosen separate Most Valuable Players (MVPs) following each regular season. From 1931, these Awards have been selected by the Baseball Writers' Association of America (BBWAA). The Baseball Almanac summarizes the early history of the Award: There have been three dierent ocial most valuable player awards in Major League Baseball history, since 1911; the Chalmers Award (1911-1914), the League Award (1922-1929), and the Major League Baseball Most Valuable Player Award [1931- ]. The MVP...is presented annually by the BBWAA. It is considered by MLB as the only ocial Most Valuable Player Award and symbolizes the pinnacle of a player's personal achievement during any single season of play. In 1938, the BBWAA began electing MVPs via a vote of the BBWAA members. Initially, there were three NL (AL) Award voters for each NL (AL) team. That number was reduced to 2 in 1961. For several decades, then, there have been 60 MLB MVP voters, where 30 participate in the NL MVP Award Election and 30 participate in the AL MVP Election following each regular season. The voting rule employed is a weighted scoring rule that has been called a (corner-weighted) version of the Borda Rule. Each voter lls in a ten-place ballot with his or her rst-place vote, second-place vote,...,and tenth-place vote. The voting system varies from a standard Borda Rule in two important respects: i) voters write in their choices (i.e., no candidates are specied on the ballot) and ii) rst-place votes receive more weight than under a standard Borda Count. Namely, players receive 14 points for each rst-place vote, 9 points for each second-place vote, 8 points for each third place vote,..., and 1 point for each tenth-place vote. Hence, a rst-place vote has the value of a second and sixth place vote rather than that of a second and tenth place vote. Under this system, the player candidate with the highest total number of points for a given league-year is crowned MVP of that league-year. Below is a valuation plot for this weighted scoring rule. Figure 1: MVP Vote Points for Each Vote Place ∗[email protected] [email protected] [email protected] [email protected] 1 A standard Borda Count system would maintain a linear trend throughout.1 Under this rule, there are (14 + 9 total points for a given League race, where a player can score as many as points.2 Σi=1i) · 30 = 1; 770 30 · 14 = 420 Baseball is a game of specialization such that cross-positional value comparisons have an apples-to-oranges quality, especially in the absence of a comprehensive, cross-positional measure of (win) value. Consider the comparison between an ace (starting) pitcher and a top shortstop. The shortstop accumulates win value through an everyday mix of oense and defense. The pitcher accumulates value almost exclusively through pitching and does so about once every 6 days. Without some exchange rate mechanism, it is dicult to compare these respective contributions. And yet, MVP voting demanded just such a comparison for decades. In 2004, Wins Above Replacement (WAR) was created by Baseball Prospectus writer Jay Jae (Jae, 2004) as a cross-positional measure of player win value (above league replacement level player at position). From 2004, WAR has been calculated for all MLB player- seasons (e.g., by Baseball Prospectus and other baseball publications). Importantly, it has also been calculated retrospectively for the universe of (preserved) professional baseball. Baseball Reference (baseball-reference.com) features retrospective calculations beginning with the rst season of the rst professional baseball league (National Association of Professional Baseball Players, 1871-75) and continuing through the respective histories of the NL (1876- ) and AL (1901- ). These histories were archived using the massive, crowd-sourced data collection project retrosheet.org, which was created by University of Delaware biology professor David Smith in 1989. The concurrent collection of past and present WAR data since 2004 has created something of a natural exper- imental setting. Since 2004, MLB MVP Award voters have cast ballots largely with knowledge of player WAR values. Before 2004, voters were relatively uninformed. Given WAR's prominence (e.g., on baseballprospectus.com, fangraphs.com, and baseball-reference.com) and the institutional nature of the BBWAA (i.e., as a dened group with a well-archived website, as well as regular chapter and national meetings), Baseball Writers who vote for MLB MVP Awards each year represent informed voters relative to their pre-2004 counterparts (many of whom are an earlier version of the same person). Beyond frequent references to WAR, by Baseball Writers (see, e.g., Madden (2017, NY Daily News) or Slowinski (2010) for a summary of the measure's importance to Baseball Writers), we refer to a Google Trends time series graph of the term's search history prominence since 2004. Figure 2: WAR Interest Over Time Source: Google Trends 1Given its extra weight on rst-place votes, MLB MVP voting may have a dierent level of susceptibility to violations of Local Independence from Irrelevant Alternatives (LIIA) when ranking the top two or three candidates, where Young (1995) states that the preservation of LIIA as it pertains to top candidates may be more important than the preservation of IIA (general form Independence) in a winner-take-all election. 2In our data, the NL and AL were imbalanced in terms of number of teams from 1998 (when the Brewers moved to the NL) through 2012 (after which the Astros moved to the AL). This aected the number of ballots designated to each League during those years. We account for this in the data by re-scaling vote points to 420 points possible for each player during those seasons. 2 From this graph, we observe that interest in WAR is cyclical. It builds throughout the season, typically peaking in August or September and falling o during the playos and into the oseason. The MLB MVP vote occurs just after the close of the MLB season in September, at or near the height of annual (search) interest. Though there was some lag until the measure went public so to speak, it is important to note that the measure originated from within the Baseball Writers Association community, an active and somewhat small organization with regular chapter-level and national-level meetings. After Jae published his seminal article on January 6, 2004, the measure quickly gained traction in leading baseball periodicals such as Baseball Prospectus. As such, the measure was visible within the Baseball Writers community soon after it was developed. The present study uses the development and retrospective calculation of WAR as a natural experiment by which to ask a question that Banerjee et al. (2011) previously consider in a starkly dierent voting context. Namely, we ask whether informed voters make better choices vis-à-vis objective information about candidate quality. Whereas fans, participants, and voters may not want the MLB MVP race to boil down to a contest of highest WAR, neither do these parties likely wish for the race to be too far removed from considerations of objective player values. Banerjee et al. (2011) conducted eld voting experiments in India and found evidence that non-partisan, third-party public disclosure of incumbent legislator report cards led to a higher voter share for high-performing incumbents. Within the present voting context, the WAR measure can be thought of as similar to a non-partisan, third-party evaluation of candidates. It evaluates candidates in a manner that imposes no a priori subjective criteria and is computationally agnostic to the MVP race. Thoth and Chytilek (2018) nd that voters facing time pressure shift their information- gathering eorts from accuracy to eciency when evaluating candidates. Specically, they restrict their attention to a smaller set of policies in candidate evaluation. In the present study, MVP voters may face time pressure in that the performance dierences of top MLB players are often slight and subtle and may require viewing hundreds of hours of game footage to perceive. In a given season, an MVP could play for any of the 30 MLB teams, each of which plays a 162 game schedule. There are 2,480 MLB games in a season, and the average game length is a little over 3 hours according to Baseball Reference. As such, there are approximately 7,300 hours (more than 304 24-hour periods) of MLB regular season games each year. Faced with this massive output of game performances, MVP voters may rely partly upon time-friendly measures such as basic counting statistics observable from box scores and video highlights to evaluate players throughout the season. The remainder of the paper is structured as follows. Section II presents the data and methodology of the study. Using a set of xed (team and season) eects negative binomial (vote) count regression models3, Section III species and tests a model of informed voting. Namely, the model asks whether MLB MVP elections from 2004 relate more strongly to an objective, comprehensive, cross-positional measure of on-eld value (i.e., WAR) than do prior elections dating from 1980. That is, are informed voters making choices that allow them to more closely identify the true Most Valuable Player (rank-ordering) for each league-season? In the absence of information on player WAR, we consider specic components of player performance and player characteristics that may have become more or less important in explaining the MVP race beginning in 2004.