Module 2

This SLP is largely about a humble and functional program called G-Power -- an excellent piece of German freeware -- that does statistical power analysis.  It can do either a priori (before the fact) or a posteriori (after the fact) analyses; before the fact analyses help you define your needed sample size; after the fact analyses help you explain why you didn't find anything interesting.

Free software, like free advice, is often worth just about what you paid for it; in this case, there are some limitations.  First thing you'll notice is that this is a Windows program with a fancy GUI. Second, it is sparse !    It still has two cardinal virtues, in my opinion:  (a) it's free, and (b) it works, fast and well.  It's nice to know that there's some sweet to go with the bitter.

In this Project assignment, you'll download and install G-Power (you can take it off later if you don't want to keep it), and use it in conjunction with SPSS to do some basic power analyses for different types of tests.  Then when you get ready to do your dissertations, you'll at least know the basics of sample size; it won't get you the sample, but it may save you from bad ones.

So -- on to the Assignment.

For this assignment you are to undertake the following steps:

1. Click here to download G-Power, and follow the instructions as to how to set it up on your machine.  The G-Power tutorial is available here; it will walk you through one basic set of analyses.  Click here for explanation of the Interface Windows.
2. A newer version (2007) of the G-Power program can be found here . You may download it - free of charge and follow the guidelines and explanations included in the website.
3. Data set from the Los Angeles Unified School District that we'll be using for this assignment are attached the description and codebook plus the SPSS version of the data.

ASSIGNMENT QUESTIONS

Begin by familiarizing yourself with the data.

1. Calculate appropriate descriptive statistics for all the variables.

2. Where appropriate, calculate 95% confidence intervals for the respective population parameters, and interpret them.  You might find it useful to obtain some graphs such as  histograms, boxplots, etc.  Add appropriate explanations to your data clarifying what the statistics tell you.

3. Test the hypothesis that there is a difference between boys and girls in their grades in the Spring Math Course.  Is the hypothesis supported?

4. Run the appropriate test and give the numbers that lead you to your conclusion.  Create boxplots showing the 95% confidence intervals for the mean of each group to illustrate and support your conclusion.

5. Use G-Power to determine the actual power of your test given your results (means, SD's, and sample sizes; use alpha=.05).  What is the size of the effect you have observed?  How do you interpret the results of your analysis?  Use Daniel Soper's free Post-hoc Statistical Power Calculator (Multiple Regression) to check your results.

http://www.danielsoper.com/statcalc/calc09.aspx

6. Suppose you were going to design a correlational study in which you couldn't assume that the effect size (r ² ) was more than .05, but you want to achieve 80% power with alpha = .05.  Use G-Power to determine how large a total sample you would need to get such a level of power.  Test your G-Power conclusions against those derived from Daniel Soper's free A-priori Sample Size Calculator (Multiple Regression). http://www.danielsoper.com/statcalc/calc01.aspx

7.  It's been alleged that boys tend to be put into bilingual education more often than girls.  Using the variables BILING2 and  GENDER, construct a 2X2 contingency table to test this hypothesis.  Is it supported?  Using G-Power, determine the post hoc power of this test to detect a true result.  If it is not supported, use G-Power to determine how large a sample would have been required to find significance in a relationship with the effect size detected in this sample.

8. We have a hypothesis that grades are inversely related to the student's reliance on a language other than English; that is, the more students use languages other than English, the lower their grades are likely to be.  For some research bearing on this hypothesis, you can click here . http://latino.sscnet.ucla.edu/challenge/lstu.htm - We have scores in two subjects -- English (ENG95) and algebra (ALGEBRA), and two measures of language use -- BILING2 (based on the bilingual use variable, and scored as 0 if no bilingual capability, 1 if some bilingual capability) and N_ENGLIS (based on the English Proficiency variable, and scored as 0 if no non-English proficiency, 1 if some non-English proficiency). These last two variables are examples of what are called "dummy variables" (no, it's not a technique named for the user, as someone once commented.)  Such a variable is essentially a yes/no coding, where a value of 0 represents not having the characteristic, and a value of 1 represents having it.  They are used to transform categorical variables, which can't be used in parametric statistical techniques such as regression, into dichotomous variables, which can be so used.  For such a dummy variable, the mean represents the proportion of the sampled having the characteristic.  In this case, they are created by recoding the base variable into a new variable as described above.  For purposes of this analysis, it's not necessary to understand this idea fully, just to accept that the coefficients for the dummy variables represent their contributions, but if you want to know more, the Social Research Methods Knowledge Base is always a good reference.  http://www.socialresearchmethods.net/kb/dummyvar.htm We'll be getting to them next module, actually. Use regression to test this hypothesis.  Is it supported for English?  For Algebra?  Use G-Power (see again presentation 14 here) to determine the power of these tests.  If either or both are NOT supported, use G-Power to determine what size sample would have been required to define as statistically significant an effect of the size actually found in this sample. [The Gloss contains an additional tutorial that may help if this is confusing you.]

9. When you're done, put your findings in the form of a report using Word®.  Try to format your results so that they are  aesthetically pleasing, or at least no worse looking than the average academic article.

Price: $22.87

Solution: The downloadable solution consists of 20 pages, 287 words and 28 charts.
Deliverable: Word Document