Conceptual Question: Experiment where you analyze different skew between conditions.

[gdbasketcase1992]

Hi all,

I had a psychology stats exam yesterday and one question was:
'Present a hypothetical experimental question in which the main interest is the difference in the skew of the dependent variable for different experimental conditions.'

I understand this to mean that you have two or more experimental groups, and the skew of the DV in each group is expected to be different and you are testing this. For example you might have condition 1 where the skew is negative, and condition 2 in which skew is positive (but the means might be the same and thus don't indicate anything). I cannot for the life of me think of an experimental question that would produce this hypothesis that experimental conditions will differ on skew of the DV.

Any ideas?

Thanks for any help you can offer!

 

[Dragan]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

Well, no, what you would assume (under a null hypothesis) is that the skew of distributions (DV) associated with the different treatment conditions are equal (i.e. the same).

 

[CB]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

Well, no, what you would assume (under a null hypothesis) is that the skew of distributions (DV) associated with the different treatment conditions are equal (i.e. the same).

True, but the substantive hypothesis of interest isn't necessarily the same as the null hypothesis, so I think OP has a decent handle on the question.... the trick is answering it This is quite a challenging exam question.

My idea: Think about an intervention that is aimed at reducing income inequality....

 

[spunky]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

True, but the substantive hypothesis of interest isn't necessarily the same as the null hypothesis

i feel like i'm missing something... if your substantive question of interest is not related to your statistical hypothesis... then why are you testing it in the first place?

#justsayin

 

[CB]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

Not sayin' they aren't related, but the substantive hypothesis is hardly ever the same as the null hypothesis. Usually the substantive hypothesis people want to test - the bit predicted by their theory or whatever - is that some relationship exists in some particular direction. Then they try to reject the null in favour of this substantive hypothesis. (Which, by the by, usually isn't quite the same as the alternate hypothesis either).

Now if us psychology types actually had theories that made statements precise enough to formulate as point null hypotheses, and then we tried to falsify them, our field would be in a lot better shape, but we don't and can't....

 

[spunky]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

Not sayin' they aren't related, but the substantive hypothesis is hardly ever the same as the null hypothesis. Usually the substantive hypothesis people want to test - the bit predicted by their theory or whatever - is that some relationship exists in some particular direction. Then they try to reject the null in favour of this substantive hypothesis. (Which, by the by, usually isn't quite the same as the alternate hypothesis either).

well... this is not necessarily true. there are examples (like covariance modelling/SEM) where your substantive hypothesis of interest is
the null hypothesis (i.e. does my model reproduce the data within sampling variability?). in linear mixed models you can also try and look for a null hypothesis to help you choose a parsimonious model.

i'm not sure if my position places me on the 'purist' side of things, but I believe that your substantive hypothesis either needs to be phrased in statistical terms to be tested OR some other hypothesis that can be phrased in statistical terms needs to be posited such that it's rejection (or failure to reject) logically implies that there is evidence in favour of your theory.

let's be honest... if you can't somehow tie hypothesis testing to your research question, then why are you doing it in the first place?

Now if us psychology types actually had theories that made statements precise enough to formulate as point null hypotheses, and then we tried to falsify them, our field would be in a lot better shape, but we don't and can't....

I think we can. we're just lazy and able to get away with it. I think if psychologists got sued for millions of dollars like pharmaceutical companies do when their new drugs come out you'd see the appropriateness of research design taking a much more prominent role in the field. but as things stand now, what's the worst thing that can happen to you? that your manuscript gets rejected?.... which, i mean, it is painful... but it doesn't hurt as much as if the editors rejected your manuscript AND charge you a bill of wasting their time.

perhaps this is a new way in which Psychology could improve itself: if you did things right, you get published. if you did them wrong, you have to pay out of your own pocket.

 

[Karabiner]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

I cannot for the life of me think of an experimental question that would produce this hypothesis that experimental conditions will differ on skew of the DV.

Reaction times typically show skewed distributions.

An experimental intervention could decrease skew for example
if it shortens reaction times particularly in those with relatively
slow reaction.

With kind regards

K.

 

[CB]

Re: Conceptual Question: Experiment where you analyze different skew between conditio

i'm not sure if my position places me on the 'purist' side of things, but I believe that your substantive hypothesis either needs to be phrased in statistical terms to be tested OR some other hypothesis that can be phrased in statistical terms needs to be posited such that it's rejection (or failure to reject) logically implies that there is evidence in favour of your theory.

I don't think we really disagree on this. I think there needs to be a connection between the substantive and statistical hypotheses. Sometimes that connection is there but a little weak, which is sort of ok but not ideal:

E.g. H0: Zero correlation
H1: Non-zero correlation
Substantive hypothesis: Positive correlation.

But it's better when it's a stronger connection:
E.g. H0 = substantive hypothesis (maybe plus some auxiliary assumptions).

I think we can. we're just lazy and able to get away with it. I think if psychologists got sued for millions of dollars like pharmaceutical companies do when their new drugs come out you'd see the appropriateness of research design taking a much more prominent role in the field.

Heh. Mayybe. Pharmaceutical companies don't test theories that make point predictions either though! But yeah, we could do better, in general (will we? will we? viva la revolucion...)