Calculate QBR for each quarterback in a modeled play-by-play frame

Aggregates a modeled play-by-play frame into the per-quarterback, per-game weighted EPA components the bundled qbr_model takes, and scores it.

The five EPA components are leverage-weighted means over the plays a quarterback was involved in: every play (qbr_epa), non-fumble sacks (sack_epa), passes (pass_epa), rushes (rush_epa) and penalties (pen_epa). The weight discounts garbage time: a play taken with the game already decided (home_wp_before below 0.1 or above 0.9) counts 0.6, one taken while it is nearly decided (0.1 to 0.2, or 0.8 to 0.9) counts 0.9, and everything in the competitive middle counts fully. The EPA feeding them is clipped at -5, with a flat -3.5 charged for a fumble.

Only players who actually threw a pass are included; a running back's carries are aggregated into the quarterback's line only when that player also attempted a pass in the game.

Usage

create_qbr(play_df, qbr_model = NULL, season = NULL)

Arguments

play_df: A modeled play-by-play data frame, as returned by cfbd_pbp_data() or espn_cfb_pbp(). Must carry EPA, home_wp_before, the pass / rush / sack_vec / fumble_vec / penalty_flag play flags, passer_player_name / rusher_player_name and pos_team. spread and home are additionally needed for the spread feature.
qbr_model: Optional pre-loaded xgb.Booster. Defaults to the bundled qbr_model.ubj from the shared cfb_model_artifacts release.
season: Optional season, used for the one-hot rule era when the frame carries no season column.

Value

A data frame, one row per quarterback per game, with:

col_name	types
game_id	character
pos_team	character
athlete_name	character
plays	integer
qbr_epa	double
sack_epa	double
pass_epa	double
rush_epa	double
pen_epa	double
spread	double
era0	double
era1	double
era2	double
era3	double
exp_qbr	double

A component with no qualifying plays is NA rather than 0 – that is the missing-value branch the model was trained against. exp_qbr is NA when the bundled model or xgboost is unavailable; probabilities and ratings are never fabricated. A frame with no identifiable quarterbacks returns a zero-row frame carrying this schema.

Examples

# \donttest{
  try({
    pbp <- cfbfastR::cfbd_pbp_data(2021, week = 6, team = "Texas", epa_wpa = TRUE)
    create_qbr(pbp)
  })
#> ── QBR data ───────────────────────────────────────────── cfbfastR 3.0.0.9000 ──
#> ℹ Data updated: 2026-09-03 22:37:52 UTC
#> # A tibble: 3 × 15
#>   game_id pos_team athlete_name plays qbr_epa sack_epa pass_epa rush_epa pen_epa
#>   <chr>   <chr>    <chr>        <int>   <dbl>    <dbl>    <dbl>    <dbl>   <dbl>
#> 1 401287… Oklahoma Caleb Willi…    29   0.611       NA    0.321     2.79   NA   
#> 2 401287… Oklahoma Spencer Rat…    20  -0.556       NA   -0.381    -1.34   NA   
#> 3 401287… Texas    Casey Thomp…    39   0.153       NA    0.258    -1.28   -1.08
#> # ℹ 6 more variables: spread <dbl>, era0 <dbl>, era1 <dbl>, era2 <dbl>,
#> #   era3 <dbl>, exp_qbr <dbl>
# }