We can use a simple method to estimate pace before turnovers were tracked, which is using the numbers of filed goals and free throws attempted by a team and assuming turnovers are a constant based on the assumed league average. That gives us a pretty accurate estimate of pace – at least in the immediate years before 1974 – and from there we can estimate offensive and defensive ratings back through the 60s and early 70s.
Neil Paine over at Basketball-Reference.com used a regression method to estimate pace, but that maps onto the 37 years of turnovers we have. The only potential issue with that is that turnovers were decreasing through the 70s and early 80s — perhaps as a result of tracking them? — and basically stabilized in the last 25 years. Neil’s data assumes that teams from the early 70s and 60s turned the ball over closer to recent rates, despite a trend toward more turnovers at the time we lose track of that data. Nonetheless, his estimations as far back as the early 60s are still within ~2 possessions of using my simple method, so regardless of method the margins of error should be fairly small. It would take an extreme shift in the league turnover average, or an historically outlying team (30 TO/game or 10 TO/game) for the method to be off by any significant margin.
In other words, FGAs and FTAs give a pretty accurate picture of how fast a team plays. In this case, here’s the formula for the simple method of pace estimation:
( Team FGA’s + Team FTA’s * 0.44 ) / Games Played / Turnover Constant
I’m using 0.974 for the turnover constant. If we had reason to believe that turnovers increased more and more as we went farther back in time, then the constant should grow smaller and smaller. There’s no evidence to show this, so I use 0.974 as the constant before 1974.
NB: that this means the further back into the 60s (and 50s we go) the wider the confidence intervals will be for the estimations.
Neil’s Regression vs. The Simple Method
If we run Neil’s regression method for 1974, the first year turnovers were recorded, it estimates four NBA teams within one possessions of their actual number, with a mean absolute value of differences* of 2.28 and a standard deviation of 2.57.
*This is the mean of all the individual differences between the actual pace and the estimated pace, taken as a positive value.
Running the simple method estimates seven teams within 1 possession of their actual number, with a mean absolute value of differences of 1.40 and standard deviation of 1.56.
An even more accurate method, available from 1971-1973 only, is to average the simple method using opponent’s data as well. For example, in 1975 this produces an average error of 1.10 and a standard deviation of 1.28. But again, that’s only useful for three seasons.