stats VHL Advanced Stats S63-S66

[Motzaburger 1,590]

VHL Advanced Statistics S63-S66

 

This is going to be a long read, so put your math hats on and get ready for some boring statistical analyses.

 

A total of 431 cases (i.e., players data) was collected from S63-S66. Data was cleaned by removing the bots using the Endurance (<99), Durability (<99), and Fighting (<40) TPE stats. This cleared the way for having named and, at one point, active VHL players only in the data set.

 

Before I get into the tests, I want to say that you will probably come across some conflicting information in this article. I will do my best to explain the discrepancies as we come across them.

 

Regression Analyses

 

First, I did similar testing to that of @Eagles did previously. This test essentially compares the relationship of a given TPE stat to a specific outcome stat (e.g., goals). I want to be clear that this test does not mean this is 100% accurate.

 

To focus on some important and alternative outcome variables, I ran 5 different tests to see which TPE best predicted the following outcomes: Goals, Assists, Shots, Plus/Minus, and Shots Blocked. I know there are other stats (e.g., giveaways, takeaways, FO%) in the portals, but I didn’t think to add any of those before I cleaned the data so those can wait for another day.

 

Shots Blocked

 

The first thing to look at is the column to the far right. This p-value must be under .05 for it to be significant (i.e., if it’s above .05, there is no relationship present). If the column does not have a significant p-value, then you can ignore that TPE skill completely in this analysis. The TPE skills highlighted in yellow are the only ones that matter.

 

The next place to look is the second column named “Unstandardized Coefficients B”. The β (or Beta) value tells us if this relationship is positive or negative. For example, if we move up one standard deviation in FaceOffs, shots blocked will either go up or down depending on if there is a negative sign.

 

Lastly, the very top row (Constant) tells us if this model is accurate. Look at the p-vlaue for the Constant and this will tell you first if you should look any further. This first model is insignificant meaning these relationships mean nothing for Shots Blocked predictions. I will discuss anyway because it’s all the stats we’ve got!

 

Interpretation of Shots Blocked Relationships:

·         FaceOffs, Scoring, and Morale are all negatively related to shots blocked.

·         Defense is responsible for blocking shots

·         This model predicted 43.3% of the variance of Shots Blocked (not significant remember)

What does this mean? Let’s start with the easy ones. The more Defense your player has the more they will block shots. No theory behind this one. The more your player has in scoring, the less they will block shots. This is likely due to the focus in offensive skills rather than defensive skills. Face-offs also makes sense. If your player is tied up and taking draws all the time, they are less likely to be in the shot blocking areas. If you have higher face-offs, your likely to block less shots because of you C role. This could also translate to the C winning the face-off, therefore there will be no shots to block if your own team wins possession in the defensive end. Lastly, Morale is an interesting one. This stat is supposed to be equal for all active players and bots (I think?), but this is not the case. There are bots and active players with different Morale levels. This one is the weakest relationship of the four, but as your Morale goes up, shot blocking goes down. I don’t have a good explanation for this one – remember there are no bots in this data either so it seems that Morale plays a role in STHS even though it is supposed to be equal.

 

 

Plus/Minus

 

The model for Plus/Minus is significant and these relationships are valid for this test. I picked Plus/Minus because it is one of the only variables we have for 5-on-5 play.

 

Interpretation of relationships:

·         Puck Handling, Defense, Morale, Strength, and Experience all positively predict Plus/Minus

·         Potential negatively predicts Plus/Minus

·         This model predicted 34.0% of the variance of Plus/Minus

 

What does this mean? Well, the higher your Puck Handling, Defense, Morale, Strength, and Experience is, the higher your Plus/Minus will be. Note that Morale is once again back in play for predicting stats. Potential will lower your Plus/Minus. This one is another interesting one as it is supposed to be equal (I think?) for all VHL players – once again a reminder there are no bots in this data set.

 

Having higher Defense is easy to understand. Same with strength and puck handling – the stronger you are the less likely to be pushed around/stepped around; the better puck handling you have the less likely you will turn the puck over and will hold onto the puck more i.e., get offensive zone time. Experience is also pretty straight forward. The more experienced you are in situations the less likely you are to make a stupid decision that will result in a goal or vice versa the more likely you will put yourself in a position to score.

 

Potential makes sense if you think that a player with a lower potential must be a worse player and therefore will be scored on more. I hope someone who is reading this can inform me about Potential and Morale as STHS skills because they both vary across all players.

 

 

Shots

 

Scoring, Puck Handling, Experience, and FaceOffs are all significant and positive predictors of shots. I won’t provide an explanation for that, it’s pretty straight forward. Passing is a significant negative predictor for shots – if you pass you’re less likely to shoot.

 

This model predicted 73.7% of the variance of Shots and is the strongest model yet.

 

 

Assists

 

Assists is a big boy and has a lot of relationships. The model predicted 69.9% (LMFAO) of the variance for Assists coming in as our second strongest model.

 

First, your platers Overall and Discipline are negative predictors of Assists. Theses ones are a bit tough to explain and I will let you use your imaginations to interpret them. To me, Overall should positively predict assists, while Discipline (if it even played a role) should be a positive predictor as well. I would love to hear other people’s explanations on this one.

 

Skating, Puck Handling, Defense, Passing, Checking, and Experience positively predict Assists. Notice how Passing is the 4^th best predictor, not the first. These all make sense to me, the order of them isn’t what we’d expect but they all are pretty clear. If you’re wondering why Checking and Defense are here think about a simple example of fore-checking. A player rushes in the offensive zone and forces the opposing d-man to turn the puck over. Good defensive skills to do this and good checking. Therefore, the d-man is out of the play and the offensive player will make a pass and start a scoring opportunity.

 

 

Goals

 

Like passing, goals is pretty easy to explain. This model was significant and is the third best model, predicting 68.3% of the variance in Goals.

I don’t think I need to explain this one very much. The higher your Puck Handling, Scoring, FaceOffs, and Experience skillz are, the more you will score. If you have higher Passing the less likely you are to score a goal because you will pass rather than shoot.

 

What do all these tests mean?

A lot. It de-bunks the myth that Overall plays an important role. It doesn’t. It opens up questions about our Potential and Morale skills in STHS. Lastly, it provides a bit of a snap shot of what skills you should upgrade depending on what stats you want to earn over the span of a season.

 

I think it is important to note that this is only from 3 seasons worth of data. It is a decent sample size, and the most recent, but it still all doesn’t mean these relationships are 100%. The other thing to note is the variance of the models. For example the Plus/Minus model only predicts 34% of the variance in Plus/Minus. This means there is 66% unaccounted for from other sources. My guess is that a lot of the variance that isn’t explained is from randomness of the simulation – which is something we cannot statically account for. Lastly, I’d like to say that since the regression models only run one outcome at a time, it is a lot easier for us to see explanations on TPE. With more advanced stats tests (that weren’t working for me today), we can see how a specific TPE skills loads onto several outcomes at once. The only issue with this is the large variation on our outcome variables (some of which are 0-100%, positive or negative, or infinite). This is why I was unable to do this today. I don’t know if there is an easy way to do that – all outcome variables would need to be measured on similar scales and converting them to be that way when we have so much stats is too much work for me (aka I don’t even know how to get that done properly so I don’t violate rules of stats).

 

I hope these models make sense and please comment and tag me if you have stats questions, I’d be happy to explain and stats questions you may have.

 

BONUS STATS! YAY!

Here is something I did for shits and gigs. This is a Confirmatory Factor Analysis test usually used to to build factors in questionnaires. What I mean by that is when you fill out a survey and you to the whole 1-7 scale where you fill in the bubble on how often you agree or disagree, this test is used to group those statements you responded to, into factors for the survey. So if you’re doing a survey on job satisfaction, you may have 100 questions, and 4 factors (25 questions each). The factors could be Management or Co-workers, or something like that.

 

I plugged all the of the TPE skills into this test and basically figured out what the 3 main types of player builds are.

 

 

First Build (your classic wannabe offensive forward):

·         Emphasis on Scoring, Puck Handling, Skating, Defense, Passing, then Strength

·         Little to no Leadership TPE

 

Second Build (1^st variation of d-man and/or two-way forward with play-making emphasis):

·         Emphasis on Fighting, Defense, Skating, and Passing

·         Little to no TPE in Leadership and low levels of Experience.

 

Third Build (2nd variation of d-man and/or two-way forward but with defensive emphasis) :

·         Emphasis on Checking, Strength, Defense, Fighting, and Skating.

 

This is neat because it provides us with the stereotypical builds that we have had in the last three seasons in the VHL. It gives us insight to where people put all their TPE based on what types of player they want to be. Pair these with the regression models from earlier and we see some patterns. Maybe it’s time for users to start spreading TPE more? Maybe it’s time for more user controlled TPE skills since we all have the same types of builds. It gets you thinking on how you may be able to make your player unique for your next build as an experiment.

 

Thank you all for reading this!

 

Word Count: >1900

Edited July 28, 2019 by Motzaburger

[Nykonax 1,564]

how about penalties/hits?

[Motzaburger 1,590]

12 minutes ago, Nykonax said:

how about penalties/hits?

Hits positively and significantly predicted from Checking (big time), then puck handling, fighting. 

Discipline negative predicts Hits - makes sense. 

 

 

PIMS significantly predicted by Checking and Fighting (big time here) and then Leadership. 

Discipline negatively predicts PIMS - makes sense. 

[Rayzor_7 741]

Do this for goalie stats next?

[Gustav 6,407]

"VHL.com article"

 

Wonderful work @Motzaburger! Certainly beats what I'm working on at the moment.

 

[Motzaburger 1,590]

11 minutes ago, Rayzor_7 said:

Do this for goalie stats next?

I was thinking about it. Might have to use bots as part of the analysis because the sample size would be so so small. Would be able to figure something out though  

 

[Motzaburger 1,590]

Claiming again for week ending Aug 11th

[Motzaburger 1,590]

Claiming last time August 18th week

[omgitshim 900]

On 7/28/2019 at 3:43 PM, Motzaburger said:

 

Plus/Minus

 

Note that Morale is once again back in play for predicting stats.

 

I don't know if Morale has a true effect in the sim or not, but it tends to go up if your team is winning and go down if it is losing. In that case, having a better +/- usually means your team is winning, which should raise your Morale. Similar with Shot Blocks, teams that are worse tend to get more, so it makes sense that there is a negative relationship there. I would consider it more of Plus/Minus or Shot Blocks helps to predict Morale rather than the other way around though.

 

On 7/28/2019 at 3:43 PM, Motzaburger said:

Assists

First, your platers Overall and Discipline are negative predictors of Assists. Theses ones are a bit tough to explain and I will let you use your imaginations to interpret them. To me, Overall should positively predict assists, while Discipline (if it even played a role) should be a positive predictor as well. I would love to hear other people’s explanations on this one.

 

For Overall, I think that's more because the people with higher Overall tend to be the ones that focus more on Scoring rather than Passing. Discipline makes a little sense here since it tends to reduce your total number of hits, so it'd be the reverse of the Checking logic. Also, I don't believe many players these days have particularly high amounts of Discipline (or Fighting for that matter) so that might skew their results due to small sample sizes.