Independent t-test without raw data?

[WaterBuffEllo4195]

I'm desperately attempting to finish my Master's thesis by this Friday and graduate. I am NOT a stats or math person and have always stuggled with these subjects.

I am using secondary data for my thesis and the summaries and totals, etc. but not the raw data. The sample size is nearly 38,000 and I need to "run" an independent t-test and include it in my thesis.

Is there a way I can calculate Levene's test, t, the degree freedom, sig, mean difference, etc. without the raw data?

Many thanks in advance!

[SmoothJohn]

I would think that talking to your supervisor would be the first thing to do.

[WaterBuffEllo4195]

Unfortunately, I'm afraid he'll tell me that I won't be able to graduate until December then.

[Dason]

I'm not sure if you can do levene's test directly (depending on which version you want to use) without the raw data but just using the summary statistics you should be able to do an independent t-test directly. Plus you could just use a Satterthwait correction and use an unequal variance t-test so you wouldn't need to worry about checking equal variances.

But on a side note I would completely get rid of the idea of faking anything ever. It's unethical and if anybody ever found out about it then you can kiss your career goodbye.

[CowboyBear]

I agree with both of Dason's points.

Not being able to do a Levene's test isn't necessarily a big deal - as Dason says you can use a Welch's t-test (t-test for unequal variances) so as to do away with testing the assumption of equal variance. In fact it's probably best practice just to use the Welch's t-test always and forget about preliminary testing of equal variances - see A note on preliminary tests of equality of variances (Zimmerman, 2004). You will just need subsample means and variances.

Faking results is not ok, ever. Hopefully what you were intending to mean with that statement isn't as bad as what it comes across as. I do also wonder about the wisdom of running an analysis in the same week that you're planning to submit.... you're not really giving yourself enough time to properly incorporate whatever the results are in your discussion, surely?

[Jake]

Can you tell us exactly what all information you do have about the data?

And in case Dason and CB haven't made it clear enough yet, you won't find anyone on this forum who will assist you in any way fabricating results/data. So just put that out of your mind.

[WaterBuffEllo4195]

I have searched the internet and library for hours and hours with no luck. I am using secondary data with a sample of 37,926. When using an onine calculator, I get a t value of -353.5194. In the research I've been doing, I've never seen a t value that large, could this even be correct?

[Dason]

It's very possible for that to be correct. With a large sample size it's very easy to find even tiny differences. If you have a moderate effect size then with your sample size you'll get a large t value since the standard error of the difference will be very small.

[hlsmith]

Are you a Political Science major? Yeah the faking it line made my heart skip a beat. You can do it in SAS using the following syntax. Post your aggregated data and we can try to help you not fake it. So please provide the Ns, Means, and Stds. FYI, posting on the internet is forever, so is publishing bogus results. Funny thing is after clicking on submit post and reviewing my typed message, the above advertisement banner now has an ad for Romney.

data test;
input Category _TYPE_ _FREQ_ _STAT_$ YourVar;
datalines;
1 0 6 N 6
1 0 6 MEAN 371
1 0 6 STD 199
2 0 6 N 6
2 0 6 MEAN 203
2 0 6 STD 35
;
run;
proc ttest data=test;
class category;
var YourVar;
run;