Statistics for BA. Help!

[flanagan10]

---

Members don't see this ad.

Hey Guys,

I am working on my BA paper, and I was hoping that someone could help me just briefly on a statistics question. I realize this is probably really elementary to a lot of you, but I'm young and learning still.

My study basically is this:

I ran males and females in a study and showed them faces of various emotional expressions. I've run a bunch of analyses, but I want to really just see if overall, males are better at labeling male faces; females are better at labeling female faces. My study already showed that overall, males are better than females at the labeling of this facial data set. But now I want to look at it based on gender. I have a column in my SPSS sheet that shows the total male faces correct, and another that says total female faces correct. And then I have a column, obviously, for gender (male = 0 , female = 1).

How can I run the analysis? If you could tell me what analysis to do, and suggest how to do it in SPSS, I'd be appreciative.

 

[clinpsychgirl]

Let's see if I actually learned anything from applied regression analysis thus far:

To keep it simple, I'd first do your basic descriptives and a correlation matrix, with gender and correct number of female faces then gender and correct number of male faces (you'd really run all of these at once, but these would be the columns you'd be most interested in). If there's anything interesting there, you can then run a regression analysis (gender being the predictor variable, and correct number of male faces and correct number of female faces being the dependent variables). From looking at the data and output you should be able to ascertain which gender is related to which outcome.

Hey Guys,

I am working on my BA paper, and I was hoping that someone could help me just briefly on a statistics question. I realize this is probably really elementary to a lot of you, but I'm young and learning still.

My study basically is this:

I ran males and females in a study and showed them faces of various emotional expressions. I've run a bunch of analyses, but I want to really just see if overall, males are better at labeling male faces; females are better at labeling female faces. My study already showed that overall, males are better than females at the labeling of this facial data set. But now I want to look at it based on gender. I have a column in my SPSS sheet that shows the total male faces correct, and another that says total female faces correct. And then I have a column, obviously, for gender (male = 0 , female = 1).

How can I run the analysis? If you could tell me what analysis to do, and suggest how to do it in SPSS, I'd be appreciative.

 

[flanagan10]

Do others agree this is the way to go?

I ran it in SPSS by doing

Analyze---> REgression---Linear

Dependent: Total Face Male

Independent: Gender

it outputs with a model summary, an anova, and coefficiants--which do i use?

 

[estrella]

Hey Guys,

I am working on my BA paper, and I was hoping that someone could help me just briefly on a statistics question. I realize this is probably really elementary to a lot of you, but I'm young and learning still.

My study basically is this:

I ran males and females in a study and showed them faces of various emotional expressions. I've run a bunch of analyses, but I want to really just see if overall, males are better at labeling male faces; females are better at labeling female faces. My study already showed that overall, males are better than females at the labeling of this facial data set. But now I want to look at it based on gender. I have a column in my SPSS sheet that shows the total male faces correct, and another that says total female faces correct. And then I have a column, obviously, for gender (male = 0 , female = 1).

How can I run the analysis? If you could tell me what analysis to do, and suggest how to do it in SPSS, I'd be appreciative.

It sounds like your hypotheses are:
1) Males are better at labeling male faces than females
2) Females are better at labelling female faces than males

I think you could just perform a t-test for each hypothesis ( IV=gender; DV =mean # of correctly identified male or female faces). Use a Bonferroni adjustment (.05/2) to adjust your alpha and not have increased risk of Type I error because you have 2 DVs.

 

[flanagan10]

It sounds like your hypotheses are:
1) Males are better at labeling male faces than females
2) Females are better at labelling female faces than males

I think you could just perform a t-test for each hypothesis ( IV=gender; DV =mean # of correctly identified male or female faces). Use a Bonferroni adjustment (.05/2) to adjust your alpha and not have increased risk of Type I error because you have 2 DVs.

Is there a way to do that in SPSS?

 

[estrella]

Is there a way to do that in SPSS?

Yes:
Analyze - Compare Means - Independent samples t-test
1) test variable = # of male face hits
grouping variable = gender

2) test variable = # of female face hits
grouping variable = gender

I think you have to do the Bonferroni by hand- but I have a really old version of SPSS- so maybe that's changed.

good luck!

 

[50960]

Male-female= categorical data that requires a chi-square I think??

 

[LM02]

Do others agree this is the way to go?

I ran it in SPSS by doing

Analyze---> REgression---Linear

Dependent: Total Face Male

Independent: Gender

it outputs with a model summary, an anova, and coefficiants--which do i use?

Because you only have one IV, you can use any of these - but, minimally, you want to report the R-squared, F (take the t and square it), and the p value. The last two values should be located in the SPSS "ANOVA" table. In this table, you want to use the values for the slope and not the intercept.

The t-tests that estrella recommended are mathematically equivalent to running a simple regression with one categorical variable. You could use the Bonferroni correction (basically just .05/2 and use that as your critical value for p), but that would be a conservative approach. It's probably unnecessary for only 2 comparisons.

To answer psisci's question, this analysis doesn't require a chi-square - but you could also look at chi-square if interested in the proportion of male hits and female hits out of the total number of exposures, and if that differs by gender.

 

[PsychMode]

Edit: Someone answered it already. 🙂

 

[homoscedastique]

Male-female= categorical data that requires a chi-square I think??

I agree. You got 2 categorical variables so i think go chi square.

Analyze--Descriptive Stat--Crosstabs
Click on one var (eg. face gender) to be row variable, click on arrow to move to box marked row.
Click on the other var (rater gender) to be the column variable as above.
Click on stats, choose chi-square, click on continue.
click on cells
In the counts box, click on observed and expected.
In the percentage section, click on rows, column and total. Click on continue and then ok

good luck!

 

[PsychMode]

I agree. You got 2 categorical variables so go chi square.

Analyze--Descriptive Stat--Crosstabs
Click on one var (eg. face gender) to be row variable, click on arrow to move to box marked row.
Click on the other var to be the column variable as above.
Click on stats, choose chi-square, click on continue.
click on cells
In the counts box, click on observed and expected.
In the percentage section, click on rows, column and total. Click on continue and then ok

good luck!

The dependent variables (# of male/femal face hits) consist of ratio data, not categorical data.

 

[lazure]

You should not be using statistical tests designed for continuous data - I do not recommend the use of t-tests, pearson correlations and regressions for your dichotomous data sets.
Look into logistic regression models and chi-squares. If you're set on using regressions, learn about dummy coding for categorical variables.

 

[estrella]

You should not be using statistical tests designed for continuous data - I do not recommend the use of t-tests, pearson correlations and regressions for your dichotomous data sets.
Look into logistic regression models and chi-squares. If you're set on using regressions, learn about dummy coding for categorical variables.

I dont agree with you for the following reason:
Only the IV is categorical ( male/female)
The DV is clearly continuous data (the NUMBER of times male and female faces are correctly identified)
T-tests are designed to be used with categorical IVs that have 2 levels and DVs that are interval/ratio.

 

[lazure]

I dont agree with you for the following reason:
Only the IV is categorical ( male/female)
The DV is clearly continuous data (the NUMBER of times male and female faces are correctly identified)
T-tests are designed to be used with categorical IVs that have 2 levels and DVs that are interval/ratio.

You know what - you're right 🙂 t-tests do sound appropriate....

 

[Psyclops]

This appears to be a two-tailed t-test. You are trying to see if there is a difference between the correct # of male faces and female faces that participants get correct. Then you want to see the results by gender. So what you do is go into SPSS under the data dropdown menu, pick split data, and then tell it to give you the output by gender. then run your t-test, it should be a repeat measures, or dependant t-test, because each participant has a score for male faces and a score for female faces. And you are comparing the mean of the # of male faces correct and the mean of the number of female faces correct. I really think its that simple. This will show you if men are better at male faces than female, and if females are better at female faces than male faces. It will not show you if males are better at male faces than females are at female faces, etc. Good Luck. The more I think about it this is the way i would do it. Keep it simple.

I do think it's funny that there is so much disagreement here. And I very well could be wrong, but what a mess that we all don't agree. And people complain about lack of convergence on clinical diagnoses...

 

[clinpsychgirl]

This appears to be a two-tailed t-test. You are trying to see if there is a difference between the correct # of male faces and female faces that participants get correct. Then you want to see the results by gender. So what you do is go into SPSS under the data dropdown menu, pick split data, and then tell it to give you the output by gender. then run your t-test, it should be a repeat measures, or dependant t-test, because each participant has a score for male faces and a score for female faces. And you are comparing the mean of the # of male faces correct and the mean of the number of female faces correct. I really think its that simple. This will show you if men are better at male faces than female, and if females are better at female faces than male faces. It will not show you if males are better at male faces than females are at female faces, etc. Good Luck. The more I think about it this is the way i would do it. Keep it simple.

I do think it's funny that there is so much disagreement here. And I very well could be wrong, but what a mess that we all don't agree. And people complain about lack of convergence on clinical diagnoses...

With regards to T-test vrs. ANOVA, I don't think there's disagreement per say...as mentioned by another poster, running two t-tests is mathematically equivalent to running a simple regression (ANOVA) with one categorical variable. And then you only run one analysis. I say its the easiest way to go. But then again, I am a relative newbie to regression analysis.

 

[LM02]

I dont agree with you for the following reason:
Only the IV is categorical ( male/female)
The DV is clearly continuous data (the NUMBER of times male and female faces are correctly identified)
T-tests are designed to be used with categorical IVs that have 2 levels and DVs that are interval/ratio.

Exactly - and a simple regression with one dummy coded categorical IV and a continuous DV is the equivalent of an independent groups t-test.

To respond to the other posts, you can always run regression models with categorical independent variables - you just dummy code or contrast code the variables. For categorical variables with more than one level, the codes have to be orthogonal.

If you have categorical dependent variables (in this case you don't) - it would be a logistic regression.

 

[flanagan10]

Haha. Wow. Good thing I don't have a psychological problem, I probably would have sixteen diagnoses by now.

So I wanted to try to do the t test, as suggested above by one of teh posters, but when I go to do it it asks me to define groups. It says "use specified values" or I can 'use cut point".

Which do I do, and what do I put in? Could anyone just explain what that means?

 

[lazure]

use specified values based on how you code males and females.

 

[Psyclops]

I have version 12.0, but you should be doing a paired-samples t-test. it will ask you to pick the two variables that should be compared. Pick the total male faces and total female faces. You should already have split the data to come out by gneder, and you are good to go. Ask againn if you have any problems.

 

[estrella]

Exactly - and a simple regression with one dummy coded categorical IV and a continuous DV is the equivalent of an independent groups t-test.

To respond to the other posts, you can always run regression models with categorical independent variables - you just dummy code or contrast code the variables. For categorical variables with more than one level, the codes have to be orthogonal.

If you have categorical dependent variables (in this case you don't) - it would be a logistic regression.

Why not keep it simple and run two t-tests? Is there any advantage to running the analysis as a regression instead of a t-test? Seem like a lot more effort to get the same result.

 

[Psyclops]

I really think you should be running a paired samples t-test. Each participant is giving you two scores. One for male faces and one for female faces. Thus, the test you run has to take into account that there is a relationship between the two sets of data. Now, SPSS will also do us the favor of splitting the data on the output into males and females. Very simple analysis, which is a good thing.

 

[LM02]

Why not keep it simple and run two t-tests? Is there any advantage to running the analysis as a regression instead of a t-test? Seem like a lot more effort to get the same result.

I guess, in my mind, a regression is just as simple as a t-test. Really, they are just different names for the same analysis. For convenience, I would go with how the data are entered into the database - if everything is already coded as 0/1, I'd do the regression. If they are entered as M/F, I would go with the t-test. (fanagan, if they're asking you to define groups, and you have them entered as M/F- one group is M and the other group is F;similarly, your groups could be 0 and 1).

But if you decide after an analysis that you want to control for additional variables, it's much easier to copy and paste your regression syntax and just add in those variables. That's why I almost always stick to regression.