Totals versus rates
Passing yards is the most quoted number in the sport and one of the least informative. A quarterback on a losing team throws more because his team is behind, so a large total can be a symptom of a bad afternoon rather than a good one.
Yards per attempt fixes most of this by dividing by opportunity. It asks how much a team gains each time it tries, which is closer to what anyone actually wants to know. Completion percentage adds accuracy but has its own bias: an attack built on short throws inflates it without gaining ground, so the two have to be read together.
Running statistics have the mirror problem. A team with a large lead runs the ball to burn clock, accumulating carries and yards in a situation the defence has conceded. Rushing totals often describe the scoreline rather than the running game.
Situational numbers
Third-down conversion rate is the most useful simple statistic available, because it measures the moment a drive lives or dies. A team that converts consistently keeps its defence rested and its opponent's attack off the field, and the compounding effect over sixty minutes is substantial.
Red-zone efficiency — how often a team turns a possession inside the opponent's twenty into a touchdown rather than a field goal — separates teams that move the ball from teams that score. The difference between four and seven points, repeated three times a game, is the difference between a good season and a mediocre one.
Time of possession is quoted constantly and is close to worthless in isolation. Holding the ball is only valuable if you do something with it, and fast-scoring attacks routinely win while holding the ball for less than half the game.
Turnovers and small samples
Turnover margin correlates with winning more strongly than almost anything else. The trap is that it is far less repeatable than it looks: fumble recoveries in particular are close to random, and a team that wins the turnover battle heavily one season usually does not the next.
This is where efficiency models earn their keep. Measures that assign a value to each play based on the situation — down, distance and field position — describe an attack far better than yardage, because they know that four yards on third-and-three is excellent and four yards on third-and-twelve is nothing.
The broader caution is sample size. Each team gets roughly a dozen drives a game and plays a short season. That is a small amount of evidence by the standards of other sports, and it is why confident conclusions after three weeks are almost always premature.