im so cnfused about independent t test

[heaven_66au]

Please someone explain to me this so I can understand;

I am working on an independent t test and no matter how much of the text book I read, or how many videos I watch I just dont understand what exactly the output from my independent t test is telling me;

So I am looking at depression rates between males and females.

They have all answered a questionaire that rated their depression on how often they felt certain things over the past week such as 'I felt hopeless' on a 0-3 scale with 0 being never, 1 being sometimes, 2 being often and 3 being all the time kinda thing.

So I have run an independent t test and it is telling me that there is a significant difference between the males and females BUT what exactly is this telling me????

Is it telling me that more women suffer depression (e.g 10 women did while only 2 men did) OR is it telling me that on average women suffered depression at a level 2 rate while men only suffered it on average of a level 1 rate?

so is it telling me that MORE women suffer depression or that women suffer depression MORE intensely??

Thank you sooooo much to whoever can help me with this. I think I have just been thinking about this too hard and have now confused myself!!

 

[dancheeee]

Hi,
according to my very modest knowledge of statistics, the direction of the relationship you can decide by running one-tailed vs. two tailed test.

In other words, if you run 2-tailed test, then you are only testing whether there is difference in females and males suffering from depression, but with 1-tailed, you can indeed test whether the depression of males is different (greater) from females and vice versa.

But what concerns me is how could you run independent t test when your data are not interval scale? That is one of the assumptions of t-test and also normal distribution of your variable I believe.

Cheers,
Danica

 

[yaklow]

Hi,
according to my very modest knowledge of statistics, the direction of the relationship you can decide by running one-tailed vs. two tailed test.

In other words, if you run 2-tailed test, then you are only testing whether there is difference in females and males suffering from depression, but with 1-tailed, you can indeed test whether the depression of males is different (greater) from females and vice versa.

But what concerns me is how could you run independent t test when your data are not interval scale? That is one of the assumptions of t-test and also normal distribution of your variable I believe.

Cheers,
Danica

Danica has explained about the directionality of the difference: ina two tailed test you look at two zones of rejection one at either end of the distribution each zone representing alpha/2 (so for 95% confidence level, that's 2.5%) while the one tailed test (a stronger hypothesis) has a zone of rejection only at one side of the distribution - you determine which that is.

Of course, in simple every day terms you probably already know the value of mean(x) and mean(y) so you know how they differ! Your test tells you whether that difference is likely due to chance given the sample size.

But...Danica is right. You ought to consider whether the Mann-Whitney U/Wilcoxon Rank Sum test is not more appropriate if you don't have a scalar dependant variable.

Ya'akov

 

[Link]

You shouldn't be doing the t-test if you only have 3 possible values that can be assigned to each person. Do a chi-square independence test if you want to determine whether there's a difference between males and females.

 

[mimestra]

Im sorry...I dont have only 3 levels that they can be.....their stress level is varied from 1-30 plus....

while they answered the questionaire according to a 1-3 point scale, the data I have is their TOTAL score...so some people had 20 some 11 some 3 some 4 some 7 etc etc.

The data has not been grouped into classifications.

So, is an independent t test right test or not?

 

[yaklow]

The t-tests (there are more than one) compare means: they tell you for example whether two groups (blonds and brunettes say) have the same *mean* score on the dependent variable or not - that's the two tailed test: its h0 is mean x1 - mean x2 = 0. You can also test the same group on repeated scores or on different scores to see if there is a difference - again you're comparing the means of the dependent variable. Your conclusion is always about the mean of one group compared to the mean of another or the mean of a group on one score agains the mean of that group on another.

As you describe the data I am not clear that the dependent variable is interval data. If it's not at least an interval scale variable then you should not use Student's t-test. To compare scores on a ranked variable for two groups there is a non-parametric test you can use - Mann-Whitney's U test.

Your data are interval data if you can say that the distance in your domain between a score of ten and a score of twenty is the same distance as between a score of twenty and a score of thirty. Thus, the temperature scale is an interval scale (at least) but the likert scale used in questionnaires is not (we don't know that my difference between "agree" and "strongly agree" is the same as yours or as between "agree" and "neither agree nor disagree"). You just can't use the t-tests with non-scalar variables.