{"metadata":{"kernelspec":{"name":"ir","display_name":"R","language":"R"},"language_info":{"name":"R","codemirror_mode":"r","pygments_lexer":"r","mimetype":"text/x-r-source","file_extension":".r","version":"4.0.5"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Field Goal Unit Analytics - Optimal Holder Positioning","metadata":{}},{"cell_type":"code","source":"library(\"IRdisplay\")\ndisplay_png(file=\"../input/nfl-big-data-bowl-2022-media/header.gif\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:03.539131Z","iopub.execute_input":"2022-01-06T17:34:03.542522Z","iopub.status.idle":"2022-01-06T17:34:03.969579Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"<a id=\"summary\"></a>\n# 1. Summary\n\nIn the analysis below, new metrics are defined using \"Next Gen Stats\" tracking data focused on field goal unit positioning and execution. These innovative stats can help quantify skill in positions that are traditionally difficult to measure (i.e. long snapper and holder). They can also be used to evaluate the optimal strategy for positioning the holder. This can be particularly interesting on long field goal attempts where the optimal strategy must balance tradeoffs between: (1) being closer to the line of scrimmage — putting kicker closer to the punt block defense; vs. (2) being further from the line of scrimmage — forcing a longer kick.\n\nThe results of this research suggest that holders tend to position themselves further from the line of scrimmage on longer kicks, but this may not be optimal — as **positioning the holder further from the line unintuitively increases the likelihood of a blocked kick**.\n\nAnecdotal evidence suggests this can be explained by at least the following two factors:\n\n1. After controlling for kick location, kickers tend to position themselves consistently from the line of scrimmage, no matter the holder location. This means that when the holder is positioned further back, the **kicker will often start out closer to the holder than normal**, which can throw off timing. [Kicker coaching guides suggest](https://fnfcoaches.com/coaching-the-placekicker-measuring-the-steps/) that the distance between placement foot and the ball is how kickers control lift. While not proven as part of this study, it is plausible that lining up closer to the holder can be leading to subtle changes in how the foot is striking the ball and contribute to **kicks being angled lower** — which would increase the likelihood of being blocked.\n1. With the holder positioned further from the line of scrimmage, **edge defenders can attack with a more direct angle** to the kicker (i.e. run around blockers instead of through). Additionally, the extra distance means the **snap will take longer as the ball has to travel a longer distance** to the holder. This results in the defense having more time to rush the kick (unless the holder handles the snap faster than normal). The combination of the defense having more time and a more direct path to the kicker can be a recipe for disaster for the offense — resulting in an increased likelihood of the kick being blocked.\n\nFuture research on this topic should incorporate vertical ball tracking data (i.e. z-axis — not currently provided in the Big Data Bowl dataset) which would allow for deeper analysis of kicking angle dynamics and also weather data which was not taken into account.\n\nThere are many possible strategy implications from this research (see [conclusion](#conclusion)), however the simplest takeaway may be **for holders to not change their positioning based on kick distance**.","metadata":{}},{"cell_type":"markdown","source":"<a id=\"background\"></a>\n# 2. Background\n\nWe first must understand the basic dynamics of kicking — the further the kick, the more likely the kick will miss and/or get blocked.","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-images/by_distance.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:03.973788Z","iopub.execute_input":"2022-01-06T17:34:04.020079Z","iopub.status.idle":"2022-01-06T17:34:04.070810Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"*Have you ever wondered how kickers change their approach on short vs. long field goal kicks?* — [It turns out they don't](https://www.nytimes.com/2016/11/06/sports/football/football-chip-shot-field-goal-kickers.html) (or at least they say they don't):\n\n> \"I kicked a 20-yard field goal the exact same way I would kick a 50-yard field goal.\" (Jay Feely)<br><br>\"I did my best to hit every kick the same every time. It just doesn’t always work out.\" (Matt Stover)<br><br>\"You can’t change your mechanics. You want the same leg swing.\" (Nick Folk)\n\n*What about the aggressiveness of the opponent block, does that change the kickers approach?* — [It doesn't for Dustin Hopkins](https://www.sbnation.com/a/nfl-preview-2019/kickers):\n> “I have a pretty high takeoff angle, and an oncoming kick rush shouldn’t have any impact on getting my kick off...I try not to change anything. I make no adjustments...I’m not worried about the rush. If I take care of what I need to take care of — if my timing’s right and I’m not slow — and the integrity of my line is good, then they can’t really affect me either way.”\n\n*If kickers aim for consistency, no matter the situation, then what about the holder? — Do holders adjust from the typical 7-8 yards behind the line of scrimmage based on situation? Should they?*\n\nBy positioning closer to the line of scrimmage, holders have the potential to shave off an extra yard or two on long kicks. Alternatively, holders can line up further back to give the kicker more flexibility on kicking angle and also force the defense to travel a longer distance to get to the kicker.\n\nAs part of the 2022 Big Data Bowl data, we explore some of these trade-offs using Next Gen Stats (NGS) tracking data. NGS data provides valuable location information for the ball and players.\n","metadata":{}},{"cell_type":"markdown","source":"<a id=\"objective\"></a>\n# 3. Objective\n\nIn the analysis below, innovative metrics are defined to evaluate positioning and execution of the kicking unit (applicable for both field goals and extra points). The objectives of this research are to: (1) use these new metrics to learn about player behavior and execution on kick plays; (2) broadly analyze which of these metrics can influence different kick outcomes; and (3) deep dive the most optimal strategy for positioning the holder on longer kicks.","metadata":{}},{"cell_type":"markdown","source":"<a id=\"methodology\"></a>\n# 4. Methodology\n\n## Data\n\nThis research was based on analyzing 5,955 field goal and extra point kick plays from the 2018-2020 NFL seasons. Note that ~1% of plays were filtered from the provided data set because of data quality issues i.e. no tracking data, no kicker, missing snap/kick event, ball location data quality issues (see [Appendix A.1](#a.1)), etc.\n\n## Play Events\nUsing the NGS data, each kick attempt is broken into segments, corresponding to the following events:\n1. **ball_snap** - The time that the ball was snapped (frame prior to snap ball movement)\n1. **ball_caught** - The time that the ball was caught by the holder (frame where ball decelerates)\n1. **ball_placed** - The time that the ball was placed by the holder (frame where ball stabilized)\n1. **ball_kicked** - The time that the ball was kicked (frame prior to kick ball movement)\n\nTo estimate when each of the events occurred, a [Piecewise Linear Regression](https://jekel.me/piecewise_linear_fit_py/index.html) model was fit onto vertical (x-axis) ball movement data for each play (limited to five frames after the provided ball_kicked event). As we can see in the example play below, each of the events correspond to the ball changing speed/direction. It is assumed that the most apparent shifts in direction correspond to the snap, caught, placed, and kicked events.","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-media/kick_events.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:04.073470Z","iopub.execute_input":"2022-01-06T17:34:04.075043Z","iopub.status.idle":"2022-01-06T17:34:04.112921Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Metrics\n\nUsing NGS player/ball location data and play events (described above), the following metrics are defined:\n* **Snap Positioning:**\n * **holder_distance** - The (x-axis) distance between the holder and the ball\n * **kicker_distance** - The (x-axis) distance between the kicker and the holder\n* **Execution Time:**\n * **snap_to_caught** - The amount of time between ball_snap and ball_caught events (how fast the snapper snapped the ball)\n * **caught_to_placed** - The amount of time between ball_caught and ball_placed events (how fast the holder handled the ball)\n * **placed_to_kicked** - The amount of time between ball_placed and ball_kicked events (how long the ball was placed before it was kicked)\n * **snap_to_kicked** - The amount of time between ball_snap and ball_kicked events (sum of three prior components)\n\nThese metrics can be used to help us learn innovative insight about kicking plays. For example, this data can help us understand how the snapper/holder impact kick success, which players excel in different aspects, how strategy/approach adapts based on the situation, etc.","metadata":{}},{"cell_type":"markdown","source":"<a id=\"results\"></a>\n# 5. Results\n\n## How do positioning and execution vary based on kick outcome?\n\n*Below we can see trends of how the metric vary based on kick outcome. The distributions are normal with minimal outliers (after data quality filters) so simple averages are being compared.*\n\nBlocked kicks tend to be situations where:\n* Holder lines up further back\n* Kicker lines up closer to the holder\n* Holder is faster at handling the snap\n* Kicker is slower to kick the ball after placed","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-images/metric_by_outcome.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:04.115715Z","iopub.execute_input":"2022-01-06T17:34:04.117384Z","iopub.status.idle":"2022-01-06T17:34:04.193680Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## How do positioning and execution vary based on kick distance?\n\nOn longer kicks (>=45 yards):\n* Holder lines up further back - *Possibly allowing the kicker to angle the ball lower to get extra distance*\n\nOn longest kicks (>=65 yards):\n* Kicker lines up further from the holder - *Possibly giving the kicker a longer run to generate more power on the kick*","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-images/metric_by_distance.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:04.196528Z","iopub.execute_input":"2022-01-06T17:34:04.198302Z","iopub.status.idle":"2022-01-06T17:34:04.234694Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## How do positioning and execution impact long field goal success?\n\n**Blocked** long kicks (>=45 yards) tend to be situations where:\n* Holder lines up further back\n* Kicker lines up closer to the holder - *Since total distance is relatively neutral within the long kick segment, the kicker may be unconsciously adjusting to be a \"normal\" distance from the line of scrimmage (instead of from the holder)*\n* Snap is slower - *Likely due to having to travel a longer distance*\n* Snap is handled faster - *Holder timing may be rushed since the snap was late*\n* Kicker is slow to kick the ball - *Kicker timing may be \"off\" as not lined up normal*","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-images/long_field_goals.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T18:53:17.757406Z","iopub.execute_input":"2022-01-06T18:53:17.762675Z","iopub.status.idle":"2022-01-06T18:53:17.807020Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## What is the most optimal holder positioning?\n\nThe data above suggests that kicks where the holder line up further back are more likely to be blocked or miss short. However one potential concern with the above analysis is that better kickers will likely be given more opportunities to attempt longer kicks. Below we filter to kicks of 45-50 yards, with the assumption being that this distance is within the range of all NFL kickers and any potential selection bias should be minimized.\n\nBelow we can see this trend remains — holders should not be lining up further back on longer kicks! On plays with a kick distance of 45-50 yards, if the holder distance was >=8.5 yards, then the kick was blocked ~3.2% of the time (vs. ~0.3% when the holder distance was shorter). Similarly, kicks missed-short 6.5% of the time with the longer holder distance (vs. 2.4% with the shorter holder distance).","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-media/holder_positioning.png\")\ndisplay_png(file=\"../input/nfl-big-data-bowl-2022-media/holder_positioning_box_plot.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:42:55.326378Z","iopub.execute_input":"2022-01-06T17:42:55.328161Z","iopub.status.idle":"2022-01-06T17:42:55.360157Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Case Study: Justin Tucker\n\nIn the following two plays, Tucker is kicking similar field goals (44.9 vs. 46.4 yards), with similar execution (1.6 seconds for both). In the first play, the holder distance was 7.9 yards, whereas in the second play it was 8.9 yards. The snap was slightly faster on the first play (less distance to travel) which gave the holder an extra time to place the ball (0.7 vs. 0.6 seconds). Tucker was lined up further away from the holder on the first play (2.3 vs. 2.0 yards).\n\nIn the first play the edge rusher (#28) appeared to try to go through the blocker (#86) to get to Tucker. In the second play the edge rusher (#21) took a more direct angle towards Tucker (i.e. wasn't fazed by the same #86). The end result was a blocked kick. If the holder wasn't lined up an extra yard away this kick might not have been blocked...","metadata":{}},{"cell_type":"markdown","source":"### Holder Distance = 7.9 yards / Kicker Distance = 2.3 yards","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-media/holder_positioning_good2.gif\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:04.397159Z","iopub.execute_input":"2022-01-06T17:34:04.399163Z","iopub.status.idle":"2022-01-06T17:34:04.855610Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Holder Distance = 8.9 yards / Kicker Distance = 2.0 yards","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-media/holder_positioning_bad2.gif\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:04.858876Z","iopub.execute_input":"2022-01-06T17:34:04.860514Z","iopub.status.idle":"2022-01-06T17:34:05.039516Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Across the longest field goals by Tucker, this kick (the only blocked in this subset) had the longest holder distance (8.9 yards) and shortest kicker distance (2.0 yards).","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-media/case_study2.png\")","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T17:34:05.041892Z","iopub.execute_input":"2022-01-06T17:34:05.043252Z","iopub.status.idle":"2022-01-06T17:34:05.086630Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"<a id=\"conclusion\"></a>\n# 6. Conclusion (Strategy Implications)\n\nBelow is a list of possible ideas for how player strategy could be adapted based on this research:\n\n* **Snapper** - Snap the ball faster if the Holder seems further back\n* **Holder** - Don't change positioning on longer kicks (limit to <8.5 yards) / Experiment with lining up closer\n* **Kicker** - Don't line up too close to the Holder and consider shading further away on longer kicks\n* **Edge Rusher** - Straight line attack from the outside if the holder is positioned further back\n* **Interior Rusher** - Attack up the middle knowing a low kick might be on the way if the Kicker is closer to the holder\n* **Coach** - Call a timeout if the Holder looks too far back","metadata":{}},{"cell_type":"markdown","source":"<a id=\"appendix\"></a>\n# Appendix","metadata":{}},{"cell_type":"markdown","source":"<a id=\"a.1\"></a>\n## A.1. Data Quality Challenges\n\nIt is worth noting that the ball_snap and ball_kicked events were already provided in the NGS data, however data quality audits revealed that these events weren't always accurate (i.e. often off by a few frames from what the ball/player movement data would suggest). To improve accuracy, these events were re-calculated. However, the provided snap and kick events were used as a \"prior\" for where the event was most likely to have taken place, and in the majority of cases the provided event time was within a frame of the calculated event time.\n\nAnother notable challenge with this research was the ball movement data was observed to be of much lower quality than player movement data. Audits revealed a number of data issues including plays where the ball didn't move across the play, plays where the player and ball movement were not in sync or scenarios where the ball would suddenly jump locations (i.e. \"teleporting ball\"). The later seem to happen frequently (~20% of plays) and most commonly around the time of snap where the ball seemingly \"teleports\" to the holder which is likely a systematic issue with the data collection process. Various techniques were used to try to best account for such issues but in a small number of cases problematic plays were simply filtered from analysis.\n\n<a id=\"a.2\"></a>\n## A.2. Team Trends","metadata":{}},{"cell_type":"code","source":"display_png(file=\"../input/nfl-big-data-bowl-2022-images/team_trends.png\")","metadata":{"execution":{"iopub.status.busy":"2022-01-06T18:39:56.237386Z","iopub.execute_input":"2022-01-06T18:39:56.239133Z","iopub.status.idle":"2022-01-06T18:39:56.251159Z"},"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]}]}