you never know...
Data Analytics and Army Football

some will find the second paragraph under the picture interesting -- the one about deferring after winning the coin toss

data was skewed because we twice won the toss, chose to receive and scored on the ensuing possession, but the opponents came right back to even things up. The defense was so bad I'm not sure valid projections can be extrapolated.
     My point last year was the offense was productive, the defense porous so why not choose to receive hoping to take the lead. Typically it makes sense to defer, both in the NFL and NCAA.

Based on a sample size of 48 (or 49), the author looks at the difference in predicted (or raw) winning percentages and finds Army has a higher winning percentage in games in which it defers.  The only thing this data tells us is that regardless of the strategy invoked on the coin toss at the start of each game, Army has a losing record.

This finding is not evidence that deferring the coin toss is actually causal, so it really doesn't mean anything.  The fact that deferring is endogenous (deferring is a conscious choice by coaching staff) rather than something exogenous such as wind, temperature, etc. means this statistic tells us nothing.

This is the consequence of simply analyzing data without bringing any theoretical model to the data (e.g. Explaining the ways in which deferring might actually heighten one's likelihood of winning, such as teams that defer are much more likely to get back-to-back possessions (data shows they are more likely to have the last possession of first half and by definition can have first possession of second half; opportunity to use half-time to game plan for first offense series of second half).

This is where the analogy to Moneyball breaks down. The A's had analysts who found that own On-Base Percentage was highly significant in explaining team winning percentage (which follows directly from a production function for wins in which baserunners (and hence OBP) are an input for scoring runs, and winning requires a team to outscore its opponent), yet was not a significant predictor of a player's salaries.  Teams were still negotiating contracts based on batting average.  So, the A's realized that players with high OBPs were undervalued in the MLB labor market, and seized on the arbitrage opportunity.

There is also no model that is used to actually predict the single game outcome of wins in a multivariate context, controlling for opponent characteristics and game-day conditions.  Finding a difference in raw means does not mean that the finding will persist controlling for other observables.

The Moneyball hypothesis is theory-based, unlike much of Sabrmetrics (uninformed data analytics that "uncovers" many correlations); a point that is lost on most.

Informed data analytics has moved to the analysis of much higher-frequency data, and is being used by teams such as the Dallas Mavericks to help inform in-game decision-making by the coaching staff.

There is no claim of causality in the linked article.

I've been told that there's "more to follow..."

So writing about nebulous, nonsensical correlations is supposed to inform us how?

If it is meaningful analysis, I sure would like to hear some defense of it.

All this data makes no sense unless we had a defense . We have had no defense so whether we receive or defer is pointless.
It is better to receive the ball first as we might lead in the game for a minute and a half of us scoring before our opponent ties it up anyway.
At least leading for a minute or so makes you feel a bit better for that minute.

BeatNavy wrote: There is no claim of causality in the linked article.

BN, you've become quite a scold.  Are you off your meds again???

reddog37 wrote: So writing about nebulous, nonsensical correlations is supposed to inform us how?

If it is meaningful analysis, I sure would like to hear some defense of it.

I'm sure you're familiar with looking at data for correlations, right?  Perhaps as one tool to help identify variables for further study?

armyfan wrote:

BeatNavy wrote: There is no claim of causality in the linked article.

BN, you've become quite a scold.  Are you off your meds again???

That's one explanation.  Another might be that other folks are off of theirs.  

For purely descriptive statistics, that's fine, but it cannot be used to guide deicsionmaking.  If I find green-eyed people are 23 percent more likely to drink Pepsi than Coke, how does that "help identify variables for further study?"  Suppose Army is 12 percent more likely to beat opponents whose schools do not have an "a" or "o" in their school name; should we plan the schedule accordingly?  That's why causality matters greatly, whether it is mentioned in an article or not.  I guess I did not get the memo that forum participants were required to restrict their comments only to linked articles in a thread.

Theory suggests relations between variables, and hypotheses derived from theory can be tested on the data.  Otherwise, you are just fishing in the data, and we know that the "pattern recognition" feature in our brain's "software" is likely to "find" spurious correlations even where none exist.

RD:
To support one of your points, Stony Brook deferred to the second half and wound up running twenty eight consecutive plays between the two halves.

reddog37 wrote: For purely descriptive statistics, that's fine, but it cannot be used to guide deicsionmaking.  If I find green-eyed people are 23 percent more likely to drink Pepsi than Coke, how does that "help identify variables for further study?"  Suppose Army is 12 percent more likely to beat opponents whose schools do not have an "a" or "o" in their school name; should we plan the schedule accordingly?  That's why causality matters greatly, whether it is mentioned in an article or not.  I guess I did not get the memo that forum participants were required to restrict their comments only to linked articles in a thread.

Theory suggests relations between variables, and hypotheses derived from theory can be tested on the data.  Otherwise, you are just fishing in the data, and we know that the "pattern recognition" feature in our brain's "software" is likely to "find" spurious correlations even where none exist.

ONE TOOL to HELP identify variables for further study. Subsequent work, including the use of OTHER tools, could lead to excluding those variables from further analysis.  

FWIW, I have previously provided links to the rationale behind deferring, but some on this board still fail to understand the strategy.

BeatNavy wrote:

armyfan wrote:

BeatNavy wrote: There is no claim of causality in the linked article.

BN, you've become quite a scold.  Are you off your meds again???

That's one explanation.  Another might be that other folks are off of theirs.  

Now that's the old BN we all remember!!!!

Correlations? Variables? All this scientific data if we receive or kickoff?
All you gotta do is get better especially on our defense and all this stuff eventually translates to nothing.
With the defense we had this past year is if we kick off to start a game, we usually are behind by 7 after 3 minutes. If we take the ball to begin with, we are ahead by 7 after 3 minutes, tied up by 10 and down by 7 as the first quarter ends. We usually got outscored 2 TDs to 1 with the defense we put out on the field last year, so whats the difference whether we receive or kick off by the time the game is over? 

More to follow? What's more to follow? Enough of this scientific hooey. I don't really care whether we receive or kick off. I just want to score when we GET the ball and when we DON'T get the ball, not let the other guy score.
Do that, you can't lose!
That's my scientific data.

I think its great that Army is trying to find something that might give them an edge over bigger, faster and more athletic opponents. And lets face the facts, if they did find something that gave an edge, I sure hope we wouldn't write an article about it allowing other schools to copy us.