If Roy Moore Isn't Rock Bottom for Republicans, What the Hell Is?

[jayjayj 571]

1 minute ago, bcking said:

If one being false has no bearing on the others, than your point about it only takes one to "cast doubt" shouldn't be true since the one false statement wouldn't impact the others.

 

I agree that my assessment is a very basic one. I'm using the most simplistic statistics. You won't get any better from a statistician unless we want to make assumptions about the chances of them being co-dependent. It is much simpler and easier to assume independence.

 

I would stand by what I said earlier though. While I agree they may be co-dependent u find that far less likely than they are independent. I don't have any math to back that you, just logic and, I admit, assumptions.

 

And even if they WERE codependent I would say the chances go up as more accusations are added on. So again if the first accusations proved to be false, I'd be more worried about the others (higher risk for the "bandwagon effect"). If the 5th turns out false - how could the 5th accusations have impacted ones from BEFORE it? It would have to be the "conspiracy" explanation for that, which is probably the least likely of any scenario.

If you consider the political motivations that many people see, granted not all people of course, one accusation being false would certainly cast doubt on the others.  That doesn't have anything to do with math, it has everything to do with human thought and behavior.  Your math might be right for coin flips or monkeys in a room potentially writing Shakespeare plays, but it doesn't account for confounding issues like human behavior.

[bcking 3,157]

6 minutes ago, jayjayj said:

If you consider the political motivations that many people see, granted not all people of course, one accusation being false would certainly cast doubt on the others.  That doesn't have anything to do with math, it has everything to do with human thought and behavior.  Your math might be right for coin flips or monkeys in a room potentially writing Shakespeare plays, but it doesn't account for confounding issues like human behavior.

I've mentioned that there are potential confounders. I mentioned it several times actually. I'm very aware what impact they may have. I deal with them in my work life on a daily basis when reading scientific literature.

 

We can't estimate their impact. They exist, they will alter the model, but we can't predict by how much. We would need some rough estimate which we would be making up.

 

The only numbers we actually have are from the study on how often sexual assault allegations are false, or don't go to court. I didn't just pick 50% because it's the middle. The number of false allegations + the number of cases that don't go to court equalled something around 50%. That is conservative because I think we can all agree that at least SOME cases that don't go to court are actually true, but I was assuming they were all false.

 

So those are the only numbers we have, so we can make the best estimate possible. Again I don't remember the number but it was like 6% of them all being false. (If there are 9 accusations now it's actually 0.2%)

 

If you'd like to provide other data that we can apply to the model, I'd be happy to incorporate it. But it would have to be quantitative, not qualitative.

Edited November 19, 2017 by bcking

[jayjayj 571]

1 minute ago, bcking said:

I've mentioned that there are potential confounders. I mentioned it several times actually. I'm very aware what impact they may have. I deal with them in my work life on a daily basis when reading scientific literature.

 

We can't estimate their impact. They exist, they will alter the model, but we can't predict by how much. We would need some rough estimate which we would be making up.

 

The only numbers we actually have are from the study on how often sexual assault allegations are false, or don't go to court. I didn't just pick 50% because it's the middle. The number of false allegations + the number of cases that don't go to court equalled something around 50%. That is conservative because I think we can all agree that at least SOME cases that don't go to court are actually true, but I was assuming they were all false.

 

So those are the only numbers we have, so we can make the best estimate possible. Again I don't remember the number but it was like 6% of them all being false.

 

If you'd like to provide other data that we can apply to the model, I'd be happy to incorporate it. But it would have to be quantitative, not qualitative.

I only wanted you to caveat your math with the potential for human behavior to nullify it.  The simple math equation being used to bolster the argument that all of the accusers can't be lying isn't valid for this purpose, because it has no way for accounting for human behavior.  It is a valid math equation, but the use of it here is flawed, meaning it isn't the authoritative end all argument that it is being made out to be for this situation.

[bcking 3,157]

9 minutes ago, jayjayj said:

I only wanted you to caveat your math with the potential for human behavior to nullify it.  The simple math equation being used to bolster the argument that all of the accusers can't be lying isn't valid for this purpose, because it has no way for accounting for human behavior.  It is a valid math equation, but the use of it here is flawed, meaning it isn't the authoritative end all argument that it is being made out to be for this situation.

Human behavior wouldn't nullify it, it would be a confounders that isn't accounted for in the model. You would just have to adjust for it.

 

Having potential confounders isn't enough to just throw out a model. If a model is using the best available data, then it's the best we have.

 

I'm providing the best model possible with the available data. It is perfectly valid based on the information it's using.

 

I think you may mean that it isn't very reliable in this scenario. Which I agree. But it's the best we can muster unless you have additional data.

 

Edit: 

 

Kust to clarify I'm talking about a quantitative model. If we actually want to put a number to the risk that he is guilty or innocent.

 

Otherwise we are just qualitatively expressing "feelings" about the "potential" yada yada.

 

Qualitative assessments are important, but I also like to have a quantitative one. Even when there are potential confounders.

Edited November 19, 2017 by bcking

[jayjayj 571]

Just now, bcking said:

Human behavior wouldn't nullify it, it would be a confounders that isn't accounted for in the model. You would just have to adjust for it.

 

Having potential confounders isn't enough to just throw out a model. If a model is using the best available data, then it's the best we have.

 

I'm providing the best model possible with the available data. It is perfectly valid based on the information it's using.

 

I think you may mean that it isn't very reliable in this scenario. Which I agree. But it's the best we can muster unless you have additional data.

Not including any potential confounders at all, despite knowing they exist, nullifies it for use in this situation.  It invalidates the use of the simplest form of the equation for this specific case, not the equation itself in general or when confounding aspects are accounted for in the math.  

 

Instead of thinking it isn't reliable in this scenario, I tend to think it isn't reasonable.  

[bcking 3,157]

Let's try to improve the model:

 

I think the Crux of the issue is in whether or not the accusations are independent or co-dependent. The model is assuming independence.

 

So what are the ways they could be codependent?

 

1. Collusion/Conspiracy - the women had interactions between each other (direct or through intermediary) that got them to all agree to like together while hiding those interactions.

 

Any estimate, preferably with data, to attribute to this? I don't believe we can come up with one. It remains a confounders, albeit one that I logically find unlikely (given the complexity)

 

2. Bandwagon accusations - each woman is acting independently, but they are "jumping on the bandwagon" and lying because other women are accusing him and they all independently don't like him.

 

I also don't think we have numbers from any literature for this. However we can TRY to "estimate" how it would work...

 

We can provide a number for the very first accusation. Say that one is 50% (based on the available literature, and again being very conservative).

 

We can then say that the risk for each subsequent accusation goes up. By how much? We have no idea.

 

Let's say 10% increase for every new accusations. So the first is 0.5, the second is 0.55 etc...

 

If we make the math easier (it's early for me) we can do 0.5 x 0.55 x 0.6 etc 

 

With nine accusations and that adjustment the risk that they are all false goes up to 3.4%

 

I'm heading out to brunch with family so won't be able to reply. I realize this is also limited in it's reliability but I am trying to adjust for at least one potential confounders. As I've said before since we have no data I made up numbers.

[bcking 3,157]

Just now, jayjayj said:

Not including any potential confounders at all, despite knowing they exist, nullifies it for use in this situation.  It invalidates the use of the simplest form of the equation for this specific case, not the equation itself in general or when confounding aspects are accounted for in the math.  

 

Instead of thinking it isn't reliable in this scenario, I tend to think it isn't reasonable.  

It does not nullify it. I'm sorry but you clearly don't understand statistics.

 

We CANT account for the confounders since we can't input them in the model. We have no quantifiable data for them.

 

That doesn't nullify a model...also again Validity means a very specific thing in statistics and you aren't using it appropriately. The model is valid, it is just limited to the information we have.

 

Sorry I have head out now though. I'm driving (not now, but to brunch)

[jayjayj 571]

2 minutes ago, bcking said:

Let's try to improve the model:

 

I think the Crux of the issue is in whether or not the accusations are independent or co-dependent. The model is assuming independence.

 

So what are the ways they could be codependent?

 

1. Collusion/Conspiracy - the women had interactions between each other (direct or through intermediary) that got them to all agree to like together while hiding those interactions.

 

Any estimate, preferably with data, to attribute to this? I don't believe we can come up with one. It remains a confounders, albeit one that I logically find unlikely (given the complexity)

 

2. Bandwagon accusations - each woman is acting independently, but they are "jumping on the bandwagon" and lying because other women are accusing him and they all independently don't like him.

 

I also don't think we have numbers from any literature for this. However we can TRY to "estimate" how it would work...

 

We can provide a number for the very first accusation. Say that one is 50% (based on the available literature, and again being very conservative).

 

We can then say that the risk for each subsequent accusation goes up. By how much? We have no idea.

 

Let's say 10% increase for every new accusations. So the first is 0.5, the second is 0.55 etc...

 

If we make the math easier (it's early for me) we can do 0.5 x 0.55 x 0.6 etc 

 

With nine accusations and that adjustment the risk that they are all false goes up to 3.4%

 

I'm heading out to brunch with family so won't be able to reply. I realize this is also limited in it's reliability but I am trying to adjust for at least one potential confounders. As I've said before since we have no data I made up numbers.

What about a combination of the two?

[jayjayj 571]

1 minute ago, bcking said:

It does not nullify it. I'm sorry but you clearly don't understand statistics.

 

We CANT account for the confounders since we can't input them in the model. We have no quantifiable data for them.

 

That doesn't nullify a model...also again Validity means a very specific thing in statistics and you aren't using it appropriately. The model is valid, it is just limited to the information we have.

 

Sorry I have head out now though. I'm driving (not now, but to brunch)

We know confounders exist and not accounting for them certainly invalidates the model.  

 

I stated previously, your model is valid, at the very simplest form of the equation.  Expanding it to account for everything, without accounting for anything, isn't valid though.

[bcking 3,157]

I'm sorry I can't continue to argue with you about Validity. You are just wrong. Confounders don't invalidate a model. They are limitations.

 

If you can provide quantifiable data for those confounders that we can use in the model, then surely not using them would make the model invalid. But that isn't the case here.

 

I have to go. Read a book on statistics. I can provide some options later if you'd like.

[jayjayj 571]

7 minutes ago, bcking said:

I'm sorry I can't continue to argue with you about Validity. You are just wrong. Confounders don't invalidate a model. They are limitations.

 

If you can provide quantifiable data for those confounders that we can use in the model, then surely not using them would make the model invalid. But that isn't the case here.

 

I have to go. Read a book on statistics. I can provide some options later if you'd like.

 

You've admitted that you aren't accounting for any confounders in your simple equation.  That invalidates the use of it for this scenario, because those confounders could potentially prove each and every accusation to be false.  We can't underestimate the power of political motivations on an individual.  Politically motivated people have done far worse than falsely accusing someone of something - those are powerful forces that can't simply be dismissed by math and cannot be accounted for in a statistics book.

Edited November 19, 2017 by jayjayj

[Dashinka 24,890]

6 minutes ago, bcking said:

I'm sorry I can't continue to argue with you about Validity. You are just wrong. Confounders don't invalidate a model. They are limitations.

 

If you can provide quantifiable data for those confounders that we can use in the model, then surely not using them would make the model invalid. But that isn't the case here.

 

I have to go. Read a book on statistics. I can provide some options later if you'd like.

Any model must list the assumptions first and if possible try to limit them.  Having assumptions does not invalidate a model, but it makes the output from that model questionable depending on the validity of the assumptions.  As you already posted, there are several assumptions in the simple model we are discussing and to create accurate output you need to look at all of them as well as potential combinations of them.  We see this in many models that are used today.  A great example are many of the models used to explain climate today.  With such a complex system as planetary climate, there is no way a single model can account for everything and so we get the models with the valid or invalid assumptions we have today.

[bcking 3,157]

4 hours ago, Bill & Katya said:

Any model must list the assumptions first and if possible try to limit them.  Having assumptions does not invalidate a model, but it makes the output from that model questionable depending on the validity of the assumptions.  As you already posted, there are several assumptions in the simple model we are discussing and to create accurate output you need to look at all of them as well as potential combinations of them.  We see this in many models that are used today.  A great example are many of the models used to explain climate today.  With such a complex system as planetary climate, there is no way a single model can account for everything and so we get the models with the valid or invalid assumptions we have today.

I've been clear on the assumptions from the beginning. It is assuming independence for each accusations.

 

As I've tried to make clear to Jay, I'm acknowledging the confounders without accounting for then because I cannot account for them. We don't have any quantifiable data that can be used to out the chances of "conspiracy" into a mathematical model.

 

The best we can do is acknowledge the confounders (which I have since the beginning). That doesn't invalidate the model. The model is valid. It just has severe limitations.

 

Unless anyone can offer a quantifiable way to incorporate the two concerning confounders (listed above), the best mathematical model we can have is one that assumes independence. That doesn't mean it is entirely accurate, but it is the best we can do if we want some sort of number assigned to the risk. 

 

Again - it is valid, it just has limitations. Validity has a very specific meaning. It doesn't mean that the model is "right" in the sense that the percent chance is 100% accurate. It just means that, given the assumptions and limitations in the data available, the model is constructed correctly. A valid model can still have serious limitations, as this one does.

 

A similar assessment can be made of some of the climate models. There are serious limitations that we don't have the information required to account for. The models are still valid, we just can't take them as the "end all, be all".

 

This is a common problem with the "lay public" getting news articles about scientific journal articles. They don't understand the nuances, and end up thinking that the conclusions are very straight forward with 100% certainty. That is never the case, even in the largest meta-analyses.

Edited November 19, 2017 by bcking

[Voice of Reason 10,298]

6 hours ago, bcking said:

I've been clear on the assumptions from the beginning. It is assuming independence for each accusations.

.

I appreciate you trying to turn a human problem into a mathematical one, but that dog won’t hunt.

 

If the women making accusations had come forth before the statutes of limitations were up, there would be a good chance for justice.  The best we can hope for now, with or without independence or counter dependence, is that everyone will have to judge, in their own minds, whether Moore and Franken are being smeared, or are actually disgusting perverts.

 

I say let them run.  If something factual or more recent comes up, and they are guilty, they can always be removed from office for illegal behavior.  Personally not a fan of “well, if there are enough claims, then they must be guilty” line of thinking.  Kind of goes totally against our justice system.  Unless we want to go back into the dark ages and start burning women who have orgasms because they must be witches...

[bcking 3,157]

2 hours ago, IDWAF said:

I appreciate you trying to turn a human problem into a mathematical one, but that dog won’t hunt.

 

If the women making accusations had come forth before the statutes of limitations were up, there would be a good chance for justice.  The best we can hope for now, with or without independence or counter dependence, is that everyone will have to judge, in their own minds, whether Moore and Franken are being smeared, or are actually disgusting perverts.

 

I say let them run.  If something factual or more recent comes up, and they are guilty, they can always be removed from office for illegal behavior.  Personally not a fan of “well, if there are enough claims, then they must be guilty” line of thinking.  Kind of goes totally against our justice system.  Unless we want to go back into the dark ages and start burning women who have orgasms because they must be witches...

I'm not saying he "must be guilty" because of a limited quantitative model. I'm saying with the quantitative data we have the chances of him being innocent are slim. Furthermore, in my opinion the potential confounders/assumptions in that model don't increase his chances enough for me to think him innocent. They certainly make it more likely, but I don't think it's enough. That, I admit, is only qualitative and just my opinion.

 

The most logical way to approach the issue is to consider the best quantitative information available. Then evaluate the assumptions and potential confounders that have no quantitative data in the best qualitative way and try to adjust the model.

 

So I start with the quantitative approach. Take that number, and then qualitatively try to adjust it for the confounders I can't factor in the model.

 

My point has always just been that based on available data, the baseline risk of every single woman's accusation being a lie is incredibly low. Now of course the risk goes up because of those confounders we can't account for. But they would have to go up substantially for it to be more likely that it is all one big lie. The best we can do is a qualitative assessment, and I just don't see those two potential confounders being all that likely.

 

Collusion/conspiracy gets harder the more people involved. The more you involve, the greater chances you are found out. Convincing one woman to lie is one thing. Two women - harder. 9 women - quite a bit harder, without any evidence of collusion to be found.

 

I admit that the "bandwagon" confounder is far more likely. But again that may account for a woman or two...but the chances that more and more women all independently decide to lie because other people have come forward? While again certainly possible, I just don't see that taking a less than 1% chance of innocence (assuming all accusations are independent) and turning it into anything close to 50% or more.

 

Finally may also disagree on what percent chance we are willing to accept. We aren't talking about executing him, or even sending him to jail. We are just talking about him stepping down from an election. If there is a >50% chance he is a sexual predator, I'd rather he stand down. You may have a different threshold.

 

In any case you may disagree with my approach all you like. You may think the confounders can play a bigger impact. But unless you have any actual data to back that up, that is just your own qualitative assessment which is the same as I've provided. 

Edited November 20, 2017 by bcking