PractiScore Math Question

[euxx]

6 hours ago, lstange said:

No, all Virginia Count and Comstock classifiers combined. Each dot is one run. X axis is high hit factor, so results for the same classifier are lined up vertically.

 

In this case, quite a few classifiers on your chart are proving my point about accuracy.

[shred]

I looked at that chart and it said "GMs shoot 90% or better of the points" to me.

 

[lstange]

On 11/19/2019 at 2:40 AM, shred said:

I looked at that chart and it said "GMs shoot 90% or better of the points" to me.

There is a substantial difference between GM-level classifier runs (hit factor greater than or equal to 95% of HHF) and classifiers shot by GMs.

 

But in both cases correlation is positive and statistically significant.

[shred]

15 minutes ago, lstange said:

There is a substantial difference between GM-level classifier runs (hit factor greater than or equal to 95% of HHF) and classifiers shot by GMs.

 

 

Well, yeah, I know a lot of GMs and even have 2 Gs myself, but the number of them that can lay down 95%+ classifiers all day, every day, any day is vanishingly small.

 

I guess I should have said "GM runs have 90+% of the points, no matter the HF"

 

[euxx]

17 hours ago, lstange said:

There is a substantial difference between GM-level classifier runs (hit factor greater than or equal to 95% of HHF) and classifiers shot by GMs.

But in both cases correlation is positive and statistically significant.

 

Hmm. I don't see how you draw that red line there.

[lstange]

4 hours ago, euxx said:

Hmm. I don't see how you draw that red line there.

Red line is linear regression (ordinary least squares, lm() function in R).

[euxx]

5 hours ago, lstange said:

Red line is linear regression (ordinary least squares, lm() function in R).

 

Eh? Of what values and what does it signify on that chart?

[lstange]

10 hours ago, euxx said:

Eh? Of what values and what does it signify on that chart?

The red regression line is fit to the data represented by blue dots. Independent variable is HHF, dependent variable is percent of points shot, observations (blue dots) are actual classifier results. The line is drawn in such a way as to minimize the sum of squares of residuals (vertical distances between the red line and blue dots).

 

Positive slope of the red regression line suggests that on average people shoot better points on fast (high HHF) classifiers, in line with my intuition. Even if you only look at GM-level results or results from GMs. To me it makes intuitive sense to accept more Charlies on distant targets, and both charts seem to tell the same story.

 

The charts may look noisy, but there is enough data to get a pretty low p value, indicating statistical significance. One can argue that the effect size is small and has little practical significance, but it still exists.

 

If you have a mix of close up and distant targets, you really don't want Charlies on the close targets. But that would be a different, more complicated story. Right now we're just looking at the main effect.

[theWacoKid]

5 hours ago, lstange said:

The red regression line is fit to the data represented by blue dots. Independent variable is HHF, dependent variable is percent of points shot, observations (blue dots) are actual classifier results. The line is drawn in such a way as to minimize the sum of squares of residuals (vertical distances between the red line and blue dots).

 

Positive slope of the red regression line suggests that on average people shoot better points on fast (high HHF) classifiers, in line with my intuition. Even if you only look at GM-level results or results from GMs. To me it makes intuitive sense to accept more Charlies on distant targets, and both charts seem to tell the same story.

 

The charts may look noisy, but there is enough data to get a pretty low p value, indicating statistical significance. One can argue that the effect size is small and has little practical significance, but it still exists.

 

If you have a mix of close up and distant targets, you really don't want Charlies on the close targets. But that would be a different, more complicated story. Right now we're just looking at the main effect.

 

You have the baseline high hit factor of some classifiers on the x-axis, you have percent of points shot on the y-axis, and you have plotted runs by shooters classified as GM?  Yet at no point have you considered the actual hit factor shot by the shooters, just the percent of available points they shot? 

 

If anything, you should look at the relationship of percent of points shot to the actual hit factor shot.  That's all that matters.  The higher HF classifiers will typically have easier targets, which will result in better points being shot on the whole, but you have no idea if those "better points" runs actually resulted in "better scores".  

 

I don't care what some random GM shot, I care what the guy with the best hit factor shot.  I'd venture a wild guess to stoke the fire that if the x-axis was actual hit factor shot the regression line would slope the other way.

[lstange]

1 hour ago, theWacoKid said:

I don't care what some random GM shot, I care what the guy with the best hit factor shot.  I'd venture a wild guess to stoke the fire that if the x-axis was actual hit factor shot the regression line would slope the other way.

If you scroll up, there is a chart that shows only 95%+ classifiers, too. With similar slope of the regression line.

 

Putting actual hit factor on X axis will not tell you anything you don't already know. Points are in the numerator of the hit factor formula, so I'm pretty sure that there'll be strong positive correlation.

 

But the question wasn't whether shooting alphas is good. The question was whether it makes sense to accept more charlies on harder targets. I'm too lazy to calculate angular size of all targets on all 78 active classifiers and think about how no-shoots affect point of aim, so I cut the corner and used HHF as a proxy instead. You can look at this data and draw your own conclusions.

[theWacoKid]

1 minute ago, lstange said:

If you scroll up, there is a chart that shows only 95%+ classifiers, too. With similar slope of the regression line.

 

Putting actual hit factor on X axis will not tell you anything you don't already know. Points are in the numerator of the hit factor formula, so I'm pretty sure that there'll be strong positive correlation.

 

But the question wasn't whether shooting alphas is good. The question was whether it makes sense to accept more charlies on harder targets. I'm too lazy to calculate angular size of all targets on all 78 active classifiers and think about how no-shoots affect point of aim, so I cut the corner and used HHF as a proxy instead. You can look at this data and draw your own conclusions.

95% plus classifiers still don't tell you WHICH scores are actually better. Speaking of that plot, what classifiers are the ones with 95% runs with <80% points? That seems worth investigating. 

 

Hit factor will tell you what you need to know because it takes into account the critical counterbalance to points. 

 

The typical wisdom is the lower hit factor the stage the more penal the dropped points and this is true. 

 

My conclusion is your plot demonstrates higher hit factor classifiers typically have more forgiving targets. And as a general rule, this is true

[lstange]

27 minutes ago, theWacoKid said:

what classifiers are the ones with 95% runs with <80% points? That seems worth investigating.

There are only eight such cases in my sample, and six of those are CM 08-02 Steeler Standards. But I don't think it makes sense to concentrate on outliers.