posted Jul 16, 2013, 5:00 AM by M. Sencer Corlu
[
updated Jul 16, 2013, 1:28 PM by M. Sencer Corlu
]
Statistics is a must course for Bilkent GSE teacher candidates; including biology, english, mathematics, and turkish majors. It is our hope that they will use their knowledge for their thesis or more importantly to improve their teaching. I designed the course with an understanding that, as Bruce Thompson says, statistics is useful for mathphobic, as well. Regression was the central focus, including its use in (nil) null hypothesis testing. In fact, the primary goal of the course was to help them understand all research is correlational, in addition to recognizing the importance of practical significance (i.e., effect sizes and, what Robert M. Capraro was very fond of, the confidence intervals). The final project of this 5week long summer term course (42 contact hours) was to conduct a secondary analysis of TIMSS 2011 data. My students were likely to be the first to do so in Turkey as TIMSS data became available in January 2013. TIMSS data needed some inadvance preparation on my part, including merging with IDB software and assigning teacher weights. Perhaps, some of these teachers will pursue on what they learnt in the course and write scholarly papers in the future.
Some of the frequently asked questions were:  I would like to focus on gender. Is this a good research idea?
 Only if you are interested in finding out why people are male or female? Dependent variable is always the focus in empirical research.
 Can I investigate Turkish students' achievement levels in biology? I want something simple; like, descriptive analysis!
 Simple or complex, it takes at least two variables to conduct research. Otherwise, it does not qualify as research.
 What should my minimum sample size be for ANOVA?
 I don't know. Did you do a power analysis?
 How can I know an instrument is reliable or not?
 You cannot. Cronbach's alpha is a measure of the internal consistency of your data, not a characteristic of a test.
 What does internal consistency really mean?
 It may be helpful to think internal consistency as the consistent ranking of the student scores across the items.
 Some people are using the ttest in analyzing Likerttype survey data. Why is this wrong?
 It is not wrong if one can use the mean to describe the location of data. Because, that's what ttest does, compare the means.
 But, you said that the mean was meaningful only when data are at least intervallyscaled.
 I estimated an effect size of 0.7. Is this any good?
 It looks like it is. Did you use Cohen's d or etasquared or what? More importantly, if other people believed d = 2; I don't think it is noteworthy, at all.
 Why is Sum of Squares (SOS) so important?
 SOS is useful because it allows the researcher to explain the unexplained.

