PDA

View Full Version : Analysis of lift numbers



Klotz
07-19-2009, 07:32 PM
At another forum we have a thread where everyone keeps track of their three powerlifting lifts, and there's a leaderboard to see who's strongest. Anyway, I took the data and analysed it, made some graphs and did some analysis.

We have about 100 men. A small minority might be using equipment, I'm not sure whom.

http://ztolk.interestingnonetheless.net/deadVsquat.PNG

http://ztolk.interestingnonetheless.net/benchVsquat.PNG

http://ztolk.interestingnonetheless.net/deadVbench.PNG

I'm no statistician, but I did a linear fit of the big three lifts against each other. They all have moderate correlation (the closer R^2 is to 1, the tighter), with bench and squat being the most closely correlated. One interesting trend is that the slope of deadlift vs squat is less than 1, meaning that as the people get stronger in the squat, their deadlift gets closer to their squat. According to the fit, the two would be equal at 718 lb.

Another thing that these show is that it's unrealistic to talk about pure ratios between lifts; there are offsets as well, and the ratios change if these are ignored.

http://ztolk.interestingnonetheless.net/totalVweight.PNG

This is a graph of PL total and Wilks versus bodyweight. Wilks is more or less constant with weight, which is expected and desired; there's a slight positive correlation. Total increases with weight, but the (linear) correlation isn't very tight.


http://ztolk.interestingnonetheless.net/wilksVtotal.PNG

This shows Wilks coefficient vs total. There is a pretty strong correlation.

Anyway, that's what I've done for now. If there are any other statistics people want (like individual lifts vs weight), let me know. I hope you find this interesting.

dynamo
07-19-2009, 07:37 PM
well your R^2 sucks for all your data, so its not valid :P pretty interesting seeing how everyone is clumped together like that.

p.s. now we have mathematical proof for when someone asks how to up their bench we can justifiably say squat more.

Klotz
07-19-2009, 07:50 PM
Yup for every pound your squat goes up, your bench press increases by 0.61 pounds. It's science.

Sean Hatley
07-19-2009, 07:54 PM
well your R^2 sucks for all your data, so its not valid :P pretty interesting seeing how everyone is clumped together like that.

p.s. now we have mathematical proof for when someone asks how to up their bench we can justifiably say squat more.

That's funny. A couple years ago on Jesse Marunde's board a kid came on there and i think he asked Jesse or just made it an open question for anyone to answer asking how he can improve his bench. Jesse chimed in and said "Squat More" and your bench will go up

geoffsherman
07-19-2009, 07:56 PM
I am a statistician and I have often wondered about how to incorporate my job into lifting; I like seeing that someone else is thinking that way too :) If you labeled the x and y, it might be a little clearer. What I did notice, is that the relationship between the bench-deadlift and bench -squat, seem to go away as you get stronger, which tells me that you approach some limit as your strength increases which makes sense. If someone had the time, you could track your workouts by day for a year and build a large regression model to determine what exercises, sets, and reps impacted which of yoru lifts. If you then tracked your sleep, what you ate, and anything else that may impact your lifting, someone could probably build a truly tailored, individual plan. I was thinking about doing that, but then we decided to have a kid which I am guessing may take up some of that free time :)

Klotz
07-19-2009, 08:14 PM
If you follow the trendlines, bench and squat meet at around 137 and bench and deadlift meet at something irrelevant like -1200. As a statistician can you tell me if the correlations are meaningful? dynamo said they weren't.

I have access to data of people posting their three lifts every time they set a PR over several years, this was just the most recent data. It's all in a big thread so it would be very tedious to refine. But it would be cool to see how each lift increases with increases in others. Once we know the correlation between lifts we know which ones cause improvement in others.

dynamo
07-19-2009, 08:21 PM
If you follow the trendlines, bench and squat meet at around 137 and bench and deadlift meet at something irrelevant like -1200. As a statistician can you tell me if the correlations are meaningful? dynamo said they weren't.

I have access to data of people posting their three lifts every time they set a PR over several years, this was just the most recent data. It's all in a big thread so it would be very tedious to refine. But it would be cool to see how each lift increases with increases in others. Once we know the correlation between lifts we know which ones cause improvement in others.

I'm not statistician I just learned the basics for engineering and for us if the R^2 is significantly lower than 1 our data is pretty useless.

dynamo
07-19-2009, 08:23 PM
I would think you would have to track lifts which compliment each other. Something like OH press and bench and squat and DL. You may find correlations between when people squat heavy and bench a few days later (like me I squat on fridays and bench on mondays) and perhaps the hormone response from friday jacks people up for their monday lifts.

Brian C
07-20-2009, 06:21 AM
Eat more train hard. Who needs graphs. Everyone is different. Graphs don't mean anything.

geoffsherman
07-20-2009, 08:18 AM
The correlation coefficient (r-squared) can range from +1 (perfect positive correlation) to -1 (perfect negative correlation). Anything over .5 or -.5 indicates a pretty strong relationship. From the numbers you provided, I would say you have found a pretty strong relationship.

Cards
07-20-2009, 08:58 AM
very creative why to use a regression analysis. as someone pointed out the r^2 isn't in the high 90's but in most cases it's not. Id say that a 70% explanation of data isn't terrible.

Hazerboy
07-20-2009, 03:07 PM
well your R^2 sucks for all your data, so its not valid :P pretty interesting seeing how everyone is clumped together like that.

p.s. now we have mathematical proof for when someone asks how to up their bench we can justifiably say squat more.

a <.9 r^2 doesn't mean that you're data isn't valid. We've done experiments in class and found a much less than .9 r^2 for something as simple as force on an object when dropped vs. its mass -- a simple test of newton's law. In fact, I don't think we were able to find a high r^2 until we used a laser gate.

My point is that good data is notoriously hard to collect, even for very simple experiments.

And so far all we've proven is that the two variables are correlated.Proving causation is harder. you'd have to take data on people who don't squat vs. their bench numbers - presumably they'd be lower. Or even better yet, take into account bodyweight. Does truly a higher squat = a higher bench? or is it a higher squat + a higher bodyweight = a higher bench (if its the latter, all you may have proven is that bigger guys can bench more). You've already done something similiar, a higher bodyweight = a higher total, now take it further.

Hazerboy
07-20-2009, 03:13 PM
Eat more train hard. Who needs graphs. Everyone is different. Graphs don't mean anything.

lawl way to be open minded man. There's plenty weightlifting can derive from science - Loui's westside principles all came from research (He modified what worked and what didn't later, but the point is that no one would have come up with ME and DE days, cycling lifts, etc. if it weren't for the original research).

geoffsherman
07-20-2009, 04:24 PM
a <.9 r^2 doesn't mean that you're data isn't valid. We've done experiments in class and found a much less than .9 r^2 for something as simple as force on an object when dropped vs. its mass -- a simple test of newton's law. In fact, I don't think we were able to find a high r^2 until we used a laser gate.

My point is that good data is notoriously hard to collect, even for very simple experiments.

And so far all we've proven is that the two variables are correlated.Proving causation is harder. you'd have to take data on people who don't squat vs. their bench numbers - presumably they'd be lower. Or even better yet, take into account bodyweight. Does truly a higher squat = a higher bench? or is it a higher squat + a higher bodyweight = a higher bench (if its the latter, all you may have proven is that bigger guys can bench more). You've already done something similiar, a higher bodyweight = a higher total, now take it further.

Actually, you could build a multi-variate regression model to determine what variables (independent variables) had an impact on the target variable (dependent variable). That would solve your problem from above and coule incorporate the many variables that are probably in play. Just keep in mind that you need 10-20 observations (i.e. people's recorded numbers) per variable to include it in the model so you would need many more people than we currently have data for.

Klotz
07-20-2009, 05:33 PM
http://ztolk.interestingnonetheless.net/liftsVweight.PNG

The correlation isn't strong, but it's there. Deadlift is by far the strongest. I find it interesting that the slope of the deadlift fit is greater than two, indicating that if you can't pull 2x bodyweight, you're a pussy.

Wow my numbers fall like right along those lines.

If anyone wants to wade through the thread and get data vs time I can send you the link. Doubt anyone wants to though.

dynamo
07-20-2009, 05:51 PM
I find it interesting that everyone above 250 in your deadlift graph is a pussy. Seems like the same 3 people keep making the heavyweights look bad. So turrible lolz.

Xellarz
07-20-2009, 11:40 PM
I have nothing really to add. But I like this.

Cmanuel
07-21-2009, 12:34 AM
Eh, with r squared values that horrible, doesn't seem like a strong correlation. Since your values for r squared are so low, we really can't deduce much from your graphs, the data points simply don't fit that particular statistical model.

I mean think about all the variables that are in play here that you are neglecting. Age of trainee, years training, diet, weight, equipment use, raw, the list goes on.

Sorry, I know where you want to go with this, but its just not going to happen with these data points.

Brian C
07-21-2009, 06:50 AM
lawl way to be open minded man. There's plenty weightlifting can derive from science - Loui's westside principles all came from research (He modified what worked and what didn't later, but the point is that no one would have come up with ME and DE days, cycling lifts, etc. if it weren't for the original research).
Just being jokingly sarcastic ;)

Reko
07-21-2009, 07:30 PM
very creative why to use a regression analysis. as someone pointed out the r^2 isn't in the high 90's but in most cases it's not. Id say that a 70% explanation of data isn't terrible.
I'd agree with this


However, we must always remember, correlation != causation :moon:

Klotz
07-21-2009, 07:39 PM
It most certainly is not.

http://www.iq.harvard.edu/blog/sss/ci700332kn00001.gif

Anyway, I'm not trying to prove anything with these graphs, I'm just trying to show the data that I have.