Ballpark Pal projections are generated through a system of models that you can read about here. The approach focuses on separating player skills from external factors such as stadium variation, weather, and batted ball luck. Mechanically, the projections come from a proprietary game simulator, which pre-executes each game thousands of times.
Game simulations are detailed processes that mimic real baseball games as closely as possible. Each game is simulated thousands of times. In each simulation, the game unfolds play by play, with outcomes determined by probability models. At the end, we aggregate the outcomes to understand the range of possible results for the game. The advantage of this approach is that it predicts every aspect of the game while naturally accounting for unmodeled factors and hidden nuances. You can read about the process in more detail here.
Projections update for each game when the most recent simulation completes. Each game can be expected to simulate once per day until official lineups are posted, at which point the final sim runs with the posted lineups accounted for. Differences between the various iterations can be due to updated information (weather forecasts, recent player performance) and randomness. Since each game simulates a finite number of times, it's normal to see slight variations between runs even with the same inputs.
Yes, game simulations account for all the various data shown on the site.
Ballpark Pal park factors are calculated at the individual batter level, focusing on how a player's performance can vary from one stadium to another. We establish a baseline performance level for each player, using the physical characteristics of batted balls without factoring in the stadium environment. Next, we introduce the variables of the specific ballpark and the day's weather conditions into the model. This adjusted model provides a more context-specific projection of player performance. Comparing the outcomes of these two models helps determine the expected impact of the park and weather on that player's performance. Park factors for entire games come from aggregating the hitter effects over the two lineups, then weighting by expected plate appearances.
Park factors are measured against the MLB average. For example, a +10% estimate for home runs indicates the combination of park and weather is expected to result in 10% more home runs for that game.
Park effects for individual games are based on the small handful of plays that are susceptible to their environment. Most outcomes occur regardless of the park and conditions. For example, a 450-foot fly ball is a home run in every stadium, while a game with few fly balls won't be helped or hurt much by strong wind. With some exceptions, park factors are typically secondary effects that shouldn't be thought of as a primary driver of the game's outcome. They show up over time but may not be visible in the results of individual games.
The Ballpark Pal matchup model estimates a probability distribution of outcomes for a batter/pitcher combination. The model is trained on a large dataset of plate appearances and uses over 100 characteristics of batter and pitcher to learn how different traits interact. You can read about it in more detail here.
DFS projections are averages from the sim results and are based on the unique scoring system that each platform uses.
Weather data is sourced from WeatherStack.