Statistics help

[Ja Di ksw]

So, this is a long shot, but does anyone know much about statistics? I've got a giant data set.

There are 4 independent variables (3 of them are T/F, one of them has 4 possible values). I want to see which of these (or their interactions) affect the independent variable. I have access to SPSS, but I'm not that good at using it.

There's more to it than that, but it's a good place to start.

Thank you.

 

[swiftaw]

What you need is a 2*2*2*4 contingency table, or you could try logistic regression

 

[Ja Di ksw]

Ok, so, I don't really know what that means

Thankfully, I've made it a little easier. There are a bunch of trials, but there's just the 4 independent variables and 1 outcome from their different combinations. SPSS won't let me run a multivariate analysis because there's only 1 dependent variable, and I don't think a one way anova is right because I have so many independent variables . . . . Not quite sure what to run

 

[swiftaw]

Ok, so, I don't really know what that means

Thankfully, I've made it a little easier. There are a bunch of trials, but there's just the 4 independent variables and 1 outcome from their different combinations. SPSS won't let me run a multivariate analysis because there's only 1 dependent variable, and I don't think a one way anova is right because I have so many independent variables . . . . Not quite sure what to run

Yes, multivariate analysis implies that the response variable is multidimensional, you only have a 1 dimensional response.

What are type your 4 independent variables? true/false, categorical, numerical?

What type is your response variable?

 

[YouArentJ]

can you just post your data set and let one of us run it though SPSS?

(and possibly use it to publish with later)

 

[swiftaw]

can you just post your data set and let one of us run it though SPSS?

(and possibly use it to publish with later)

Sounds like another academic in the house

 

[mkrishnan]

Ok, so, I don't really know what that means

If you don't know what that means, you have to get a statistician to be your collaborator and have her/him do your stats for you. This is sort of like saying you don't know too much about cars, but you have a 2004 Celica with a fuming engine and you've just broken into the mechanic's shop down the street. It doesn't really work that way....

 

[YouArentJ]

lol - something like that (former graduate student looking to become a graduate student again) - I find it difficult to explain statistics (and difficult for most to learn) without the numbers.... Even easier to start shoving them through SPSS and see what turns up!

 

[letsgorangers]

Multiple Regression?

If you don't know what that means, you have to get a statistician to be your collaborator and have her/him do your stats for you. This is sort of like saying you don't know too much about cars, but you have a 2004 Celica with a fuming engine and you've just broken into the mechanic's shop down the street. It doesn't really work that way....

That's a little harsh. I have a lot of (graduate and undergraduate) experience in statistics, but I've never done "logistic regression".

 

[swiftaw]

If you don't know what that means, you have to get a statistician to be your collaborator and have her/him do your stats for you. This is sort of like saying you don't know too much about cars, but you have a 2004 Celica with a fuming engine and you've just broken into the mechanic's shop down the street. It doesn't really work that way....

Yeah, (not aimed directly at the OP), it really bugs me that people think they can just "do" statistics, just because they took one intro stats class as an undergrad. It's like saying that you can do open heart surgery because you took a CPR course.

 

[YouArentJ]

I had some good grades in my graduate quantitative methods classes, but two years later, i need to look at my notes and books to warm up my mind grapes.

 

[mkrishnan]

That's a little harsh. I have a lot of (graduate and undergraduate) experience in statistics, but I've never done "logistic regression".

I'm sorry, I don't mean to be harsh or insult the OP. But statistics is a complex field, and it's no service to anyone if someone tries to wing it. Someone reading this thread might think they're qualified to do a diagnostic assessment for a psychiatric disorder, but (being a couple years away from licensure myself) I am not yet qualified to do that independently, and for a good reason, although I'm relatively close. But my training in diagnosing and treating psych disorders doesn't mean that I can do neurosurgery. That doesn't mean someone ought to be handing me a scalpel and asking me to do a craniectomy.

I just think there are certain fields of science that we all somehow feel we can pull an "armchair quarterback" for, and statistics is often one of them. But that leads to a lot of really poor research out there that really misrepresents statistics. So I think we should all be aware of our limitations within statistics. Just because I can do structural equation modeling, and I've even been trained to do it with dichotomous data, doesn't mean that I have any business just running a mixed effect model in SPSS without bothering to learn how mixed effects models work first, either.

 

[swiftaw]

The big problem in 'amateur' statistics is that it is common to assume that just because they (or their computer) can crunch the numbers then the answers obtained must be meaningful.

What people often forget is that every statistical method requires a certain set of conditions to be true in order for the method to be used (two of the most common conditions are normality and equal variance). The major issue lies in the fact that if these conditions are NOT true it is still possible to run through the method and obtain an answer. However, because the method isn't valid if the conditions are not true the answer obtained, and therefore any conclusions drawn from it, have little or no meaning.

Therefore, the hard part in 'doing' statistics is choosing the appropriate method for the task at hand. Since almost all methods will actually produce an answer it is up to the user (not the computer) to choose the method that is most appropriate.

For example, a common method is ANOVA (analysis of variance) but it has to be remembered that the simple ANOVA requires that all populations have a normal distribution and all have the same variance. If these conditions are not true you can still run an ANOVA and construct an ANOVA table, however any conclusions from that analysis would be meaningless.

There are many many methods, and intro courses only introduce you to the most basic and common ones. The more experience you have, the most methods you get exposure to, and therefore the more problems you can solve. Using ANOVA or Linear Regression just because you are familiar with those methods doesn't mean that they are appropriate for the given situation.

 

[YouArentJ]

You don’t really have to know how to perform a statistical operation... You just have to be able to read and interpret what it means... That's why we have SPSS to do it for us...

That said, if you don't know what the operation does... how do you know you should be using it?!

OMG, contradicting and confusing myself... This is why I'm going to be a professor.

 

[swiftaw]

You don’t really have to know how to perform a statistical operation... You just have to be able to read and interpret what it means... That's why we have SPSS to do it for us...

That said, if you don't know what the operation does... how do you know you should be using it?!

OMG, contradicting and confusing myself... This is why I'm going to be a professor.

You are correct, you don't to know the math behind what the computer is actually doing when it crunches the numbers, but you do need to know that you are using the correct methodology for analyzing that particular data.

 

[YouArentJ]

You are correct, you don't to know the math behind what the computer is actually doing when it crunches the numbers, but you do need to know that you are using the correct methodology for analyzing that particular data.

10-4... A social scientists agrees with you!

 

[swiftaw]

10-4... A social scientists agrees with you!

Wow, might be the first time that's happened

PS I'm a stats professor.

 

[YouArentJ]

Lol

 

[letsgorangers]

I'm sorry, I don't mean to be harsh or insult the OP. But statistics is a complex field, and it's no service to anyone if someone tries to wing it. Someone reading this thread might think they're qualified to do a diagnostic assessment for a psychiatric disorder, but (being a couple years away from licensure myself) I am not yet qualified to do that independently, and for a good reason, although I'm relatively close. But my training in diagnosing and treating psych disorders doesn't mean that I can do neurosurgery. That doesn't mean someone ought to be handing me a scalpel and asking me to do a craniectomy.

I understand your point about (not) being qualified, etc. I personally hope to never be an expert in any one thing -- I want to continue to learn (about cognitive neuroscience!) for the rest of my days.

You are right -- statistics is hugely complex. But if the OP is in a field (psychology!) that stresses statistics, he isn't doing himself any favors by paying someone to do the analysis for him. He'd be better off finding a very kind professor or student to help him learn. But if this is simply the end of a project and statistics isn't something he'll address in the future, then you are right about hiring a statistician.

PS-You can have SEM! I did some factor analysis this past semester -- egh!

 

[mkrishnan]

You are right -- statistics is hugely complex. But if the OP is in a field (psychology!) that stresses statistics, he isn't doing himself any favors by paying someone to do the analysis for him. He'd be better off finding a very kind professor or student to help him learn. But if this is simply the end of a project and statistics isn't something he'll address in the future, then you are right about hiring a statistician.

PS-You can have SEM! I did some factor analysis this past semester -- egh!

You're very right (about learning statistics) ... if you're in any field that uses it on a routine basis, it's an excellent investment. Even if you don't routinely use it yourself, you'll understand others' research much better. But you have to learn. The worst thing about tools like SPSS is that they provide huge assistance in doing statistics and almost no assistance in understanding it. So you have no idea ... whether you are interpreting the statistics correctly, whether the analysis is even loosely valid, whether the data is appropriate to the analysis, etc, etc.

The time to consult / collaborate is when you really need the results now, and you have plenty of your own substantive expertise that you're bringing to the table.

And SEM rocks! But SEM programs are a huge pain in the....

 

[swiftaw]

So you have no idea ... whether you are interpreting the statistics correctly, whether the analysis is even loosely valid, whether the data is appropriate to the analysis, etc, etc.

Amen!!!

What would be helpful is if SPSS (or other software) had a training mode, where every time you ran an analysis you'd get a big popup window saying "Are you sure this is the correct method of analysis? Did you check the required assumptions of this method?"

Another thing that I see all the time is the assumption that statistics only plays a role in analyzing data, rather than in also collecting data. Many times I've had people come to me wanting to analyze their data and I've had to tell them that I can't because of the way it was collected.

To quote Light, Singer and Willett,
"You can't fix by analysis what you bungled by design."

 

[letsgorangers]

You're very right (about learning statistics) ... if you're in any field that uses it on a routine basis, it's an excellent investment. Even if you don't routinely use it yourself, you'll understand others' research much better. But you have to learn. The worst thing about tools like SPSS is that they provide huge assistance in doing statistics and almost no assistance in understanding it. So you have no idea ... whether you are interpreting the statistics correctly, whether the analysis is even loosely valid, whether the data is appropriate to the analysis, etc, etc.

I agree completely. After a few classes in undergrad using SPSS, I still felt like I had no idea what I was doing. It doesn't help at all when the professor tells you exactly what procedure to use -- once you're on your own, you have no idea. I took multivariate data analysis a few semesters ago and we very rarely did anything on the computer (and when we did, we used SAS). He didn't make us do stuff by hand TOO MUCH, but he was emphatic that we knew how to. I was lucky to have him as a teacher -- he was amazing and taught us a lot in a field that is not easy.

 

[OutThere]

There's a considerable amount of elitism floating around in this thread.

There are lots of subtleties in statistics, yes, but that doesn't mean that someone can't dabble. There's not much to lose by playing around with SPSS. An undergrad intro stats course will teach the assumptions you have to make when using various tests, and most of the basics can be found with a google search. Telling someone they don't know enough to do something may be appropriate when there is something to lose (car engines, mixing chemicals, brain surgery) but the only sour outcome from playing with SPSS could be misinterpreting the results...and surely no PhD has ever done that.

I'm not denying that it is good to learn the basics...I learned everything by hand, then on my TI89, then on SPSS, but why all the hate?

 

[swiftaw]

There's a considerable amount of elitism floating around in this thread.

There are lots of subtleties in statistics, yes, but that doesn't mean that someone can't dabble. There's not much to lose by playing around with SPSS. An undergrad intro stats course will teach the assumptions you have to make when using various tests, and most of the basics can be found with a google search. Telling someone they don't know enough to do something may be appropriate when there is something to lose (car engines, mixing chemicals, brain surgery) but the only sour outcome from playing with SPSS could be misinterpreting the results...and surely no PhD has ever done that.

I'm not denying that it is good to learn the basics...I learned everything by hand, then on my TI89, then on SPSS, but why all the hate?

I wasn't hating, just in my experience people who are dabbling in statistics tend to focus on the method and overlook the necessary assumptions for performing said method. Of course, not everyone is like that, but I have seen it too many times for it to be a rare event.

If anything, the blame probably lies somewhat with the people teaching the intro stat courses, who focus a little too much on performing the method and not quite enough on understanding and testing for the underlying assumptions.

 

[OutThere]

If anything, the blame probably lies somewhat with the people teaching the intro stat courses, who focus a little too much on performing the method and not quite enough on understanding and testing for the underlying assumptions.

This may well be true...I had a righteous b*** of an intro stats teacher who made us do tons of memorization of assumptions and theoretical bits. She'd have us write out a paragraph on the assumptions and the limitations on inference associated with each problem on homework and tests. I failed a couple of tests but got it together and aced the exam. I'm finding out now, a couple of years later, that despite being evil she may have been a great teacher, as I've been able to help Econ major friends with their stats homework a few times.