[Schatze]

Stats is the big EZ. If you aren't afraid of it, and bother to look at the logic behind it, it's not too bad. Especially if you're not doing a highly technical stats course like I did. The people who failed miserably, and a lot of people do fail stats, or do poorly, are the ones who don't bother to grasp the logic and the concepts but just look at it as writing out formula.


One tailed test, Z distribution. Pretend that looks like a normal distribution, the infamous bell curve.
_______. ^ .
______._____.
____. _______ .
_ . ___________.
/__|__|__|__|__|_\
-2SD-1SD M____^

Anything left of the arrow would be below critical threshold and you would fail to reject the null. Anything right of the arrow would be more extreme than expected depending upon your significance level. I've arbitrarily placed it near where two standard deviations above the mean would be because that cuts off 96% of the distribution, so close to .05 (or would in a normal distribution). Now, if you're doing a one tailed hypothesis test, you'd want your Z score to be to the right of the arrow. That is, that your Z score is more extreme than the critical value. If it is less extreme than the critical value, you will fail to reject the null. If it is more extreme, you reject the null hypothesis and conclude that the results are significant. The critical value corresponds to a Z score with arbitrary that is given in your tables for the cutoff for the particular significance level cutoff you're interested in. So if your results are more extreme than the critical cutoff, your reject the null and you've "proven" whatever it is you're trying to prove. If the results are less extreme, you cannot reject the null and the results are not significant.

In your step 5.) you may want to add a caveat by including a "significantly" in your conclusion. It depends on how anal the prof is. There's always a chance of error, so you can't say definitively one way or the other(p. < .05 is less than 5% chance of error, .01 less than 1% etc.); however, when speaking within the context of the test, you *did* prove it at your predetermined significance level.


If you fail to reject the null, then it means there was a statistically significant difference.

In my above example, "(test type) crticial was exceeded, therefore the null is rejected. Students who wear hoodies listen to more Linken Park then the population at large."


BTW, your p value is wrong I think. Since you're testing the null it would be p. < .2, that is if the p value is larger than .2 that means the claim is false, whereas if the claim is true the p value would be less than .2, cutting off 80% of the distribution. Remember you're testing the null hypothesis which is basically the opposite of the research hypothesis. But I'm not certain what type of stats course you're doing and how the questions are phrased so you may want to double check that you're doing it right by doing the practice problems and making sure your answers are correct. I assume you're using H to signify mu.

Correlation and linear regression: correlation is easy, you'll probably be doing the pearson product moment correlation and not a non-parametric like Spearman's rho. so you'll be deriving r.

linear regression is easy too as long as you keep your concepts straight like Y^ (should be above the Y, called Y hat). Depending how in depth you'll get it'll look something like Y^ = a + b(X1)(X2)(X3), for however many predictor variables you have (the X axis is the predictor variable(s), Y the criterion variable). a is the correlation constant which is the Y intercept, where the regression line crosses the Y axis, b is the regression coefficient which is the slope of the line. Basically that allows you to predict what your expected Y value (Y^) will be for any given X, or vice versa, which falls along the regression line.

If you get in depth, you may go onto the general linear prediction rule from which most parametric statistics are derived. This includes error, i.e. how much do the predictors actually predict the Y variable. For instance, like if you wanted to predict how well a person would do in his stats class the first exam, the 3rd exam, and the amount of homework turned in accounted for 56% of determining outcome of final mark (actual example), with 44% of how well they did being unaccounted for by those predictors.

edit: the stats courses I've done have been highly technical and theoretical, rather than "understanding stats". As in, I could do the statistical calculations or examine the statistical calculations of a published paper. I used A. Aron in my 2nd year stats, chapter one was descriptive, chapter two was explaining what inferential statistics are, 3 was T, 4 was T using n > 1, 5 Z, then each chapter to a different statistical method. Hypothesis testing was part of chapter 2. This was 2nd year stats. 3rd year was non-parametric and using SPSS mainly.

Question: what sort of stats course does hypothesis testing come near the end? I realize I was taking a course for stats in a particular branch of an academic field, but I can't imagine how you would only get around to hypothesis testing near the end rather than the beginning. Was your course more about descriptive statistics than inferential statistics?