The more ways that you try to measure something, the more accurate you can be. Dr Chris Robertson, Clippd's Chief Data Scientist, is here to explain why.
Take the example of watching two golfers hitting shots on the range. One is a PGA Tour golfer, the other is a low handicap amateur. They both hit a lot of good shots, which initially makes it hard to tell which one is which.
If, on the other hand, you watched the same two golfers hitting shots on the course, where you know their intention and have a much clearer understanding of the context of each shot, it becomes much easier to assess who the better player is. This is because the on-course information more accurately reflects how well they play than the evidence of the range.
But what if you had two measurements that are equally accurate (or inaccurate)? Is there any point in keeping both?
"There is an inherent randomness associated with measurement," explains Dr Chris Robertson, Clippd's Chief Data Scientist. "Sometimes it measures exactly right but more often than not it will be wrong by a random amount. There is most definitely an inherent randomness associated with golf. We all hit shots that do not accurately represent our levels of skill. It's the game."
So what is the point of taking more than one measurement?
In Dr Robertson's example above we have two thermometers, which are both independently measuring the temperature in the same room. The two thermometers are different but measure with a similar error of plus or minus five degrees Celsius.
"If you don't care what the real temperature is and all you need is a temperature of plus or minus five degrees Celsius, it really doesn't matter," says Dr Robertson. "But that's not what we're trying to achieve. At Clippd, we're trying to get the most precise measurements that we can in order to give you the most accurate feedback on your game. Every part of it."
So what is the answer to the question posed in the slide above? It's the final option: a number between the red and blue measurements. If you want a more precise measurement of the temperature in the room, it's better to have two thermometers. "By including more information and narrowing the window that we know the temperature must be in – in this case 18 degrees Celsius plus or minus 1 degree Celsius — we get a more precise measurement," Dr Robertson confirms.
You might think standing on the range and watching the two players hit balls bears little to no relation to how good or bad they are as golfers. In fact, watching them on the range is an additional source of data that allows us to be more accurate. "Watching the two players on the range is effectively the second thermometer," explains Dr Robertson. "The first thermometer – on-course play – might be a lot more accurate, but the second thermometer, even though it's less accurate than the first, is still giving you a signal. It’s still telling you something."
As we all know, sometimes you can hit it great on the range and can't quite get things to click on the course. On other occasions, the opposite is true. But as your ability improves to hit the shots you want to hit on the range, the chances of you executing them on the course does, too. Otherwise what’s the point of practice?
The next question is, how much do you trust the second thermometer – the off-course information – compared with what you have seen on the course?
Let’s look at it another way. If you had complete tournament data for all the players competing in the Masters in the lead-up to the event, and you were then offered the chance to look at their practice data from the week before the tournament, would you discard their practice data when attempting to select a likely winner?
No, because the practice data will contain useful information. You might not let it affect your assessment as much as if you had seen a player hit the exact same shots in a tournament, but the practice data will give you an idea of where things are moving.
So how does that practice data change your perception of how well or badly they are playing? "That's where the Clippd analysis comes in," says our Chief Data Scientist. "We ingest data from all environments – tournament rounds, practice rounds, launch monitor tests, practice drills etc – to produce an accurate measurement of your skill: Player Quality. In doing so, our Player Quality algorithm develops an understanding of the interplay between the different types of activity that is completely bespoke to you."
Collecting more data allows Clippd to more accurately assess your ability. If we can more accurately assess your ability now, we can give you a better idea of whether you're getting better or not, and how quickly.
For further reading and a more detailed technical explanation of the concepts presented here please refer to 'Kalman and Bayesian Filters in Python' by Roger R. Labbe.
Take the example of watching two golfers hitting shots on the range. One is a PGA Tour golfer, the other is a low handicap amateur. They both hit a lot of good shots, which initially makes it hard to tell which one is which.
If, on the other hand, you watched the same two golfers hitting shots on the course, where you know their intention and have a much clearer understanding of the context of each shot, it becomes much easier to assess who the better player is. This is because the on-course information more accurately reflects how well they play than the evidence of the range.
But what if you had two measurements that are equally accurate (or inaccurate)? Is there any point in keeping both?
"There is an inherent randomness associated with measurement," explains Dr Chris Robertson, Clippd's Chief Data Scientist. "Sometimes it measures exactly right but more often than not it will be wrong by a random amount. There is most definitely an inherent randomness associated with golf. We all hit shots that do not accurately represent our levels of skill. It's the game."
So what is the point of taking more than one measurement?
"If you don't care what the real temperature is and all you need is a temperature of plus or minus five degrees Celsius, it really doesn't matter," says Dr Robertson. "But that's not what we're trying to achieve. At Clippd, we're trying to get the most precise measurements that we can in order to give you the most accurate feedback on your game. Every part of it."
You might think standing on the range and watching the two players hit balls bears little to no relation to how good or bad they are as golfers. In fact, watching them on the range is an additional source of data that allows us to be more accurate. "Watching the two players on the range is effectively the second thermometer," explains Dr Robertson. "The first thermometer – on-course play – might be a lot more accurate, but the second thermometer, even though it's less accurate than the first, is still giving you a signal. It’s still telling you something."
As we all know, sometimes you can hit it great on the range and can't quite get things to click on the course. On other occasions, the opposite is true. But as your ability improves to hit the shots you want to hit on the range, the chances of you executing them on the course does, too. Otherwise what’s the point of practice?
The next question is, how much do you trust the second thermometer – the off-course information – compared with what you have seen on the course?
No, because the practice data will contain useful information. You might not let it affect your assessment as much as if you had seen a player hit the exact same shots in a tournament, but the practice data will give you an idea of where things are moving.
So how does that practice data change your perception of how well or badly they are playing? "That's where the Clippd analysis comes in," says our Chief Data Scientist. "We ingest data from all environments – tournament rounds, practice rounds, launch monitor tests, practice drills etc – to produce an accurate measurement of your skill: Player Quality. In doing so, our Player Quality algorithm develops an understanding of the interplay between the different types of activity that is completely bespoke to you."
Collecting more data allows Clippd to more accurately assess your ability. If we can more accurately assess your ability now, we can give you a better idea of whether you're getting better or not, and how quickly.
For further reading and a more detailed technical explanation of the concepts presented here please refer to 'Kalman and Bayesian Filters in Python' by Roger R. Labbe.
