Hello there, I have a major concern regarding my data analysis. I am investigati

Question

Hello there,
I have a major concern regarding my data analysis.
I am investigating a possible correlation between selective demographic (age, gender, years of teaching, content area/specialization, and school type) and technographic cahracteristics (prior experience with instructional technology and LMS use, type and amount of proffesional training received on instructional tech and LMS use) and teachers' peak stage of concern on a developmental 7-stage of conern continuum.
I already gathered my data using a 35-item 7-point Likert-type questionnaire and i plan to do logistic regression. I have already gathered 345 completed questionnaires and did a post hoc power analysis for logistic regression , using the input parameters of two-tailed, z-test, alpha=.05, a total sample size of 345, which gave me a statistical power of .49, which is far below the desired 0.80. According to Gall et al. (2007), the rule of thumb for conducting multiple regression analysis is to “increase sample size by at least 15 individuals for each variable that will be included in the multiple regression analysis” (p. 360). In my case, there are 13 variables to be entered in the regression equation and based on the above rule of thumb I need 195 participants. Is the sample of 345 acceptable to proceed with logistic regression analysis? Any help is greatly appreciated.

Liza

Explanation / Answer

I think you should be fine to proceed, you are trying to think of statistics as "right or wrong" and that you somehow will get your desired .80 statistic power. The .80 just means "a very strong correlation between data variables" .50 is a weaker correlation between your variables. It is what it is.

Again, there is no right or wrong answer, your data tells you what is tells you, even if it is something you might not like. Increasing your participant count i don't believe will affect the result of the test too much. More will likely simply tell you more of what you already have. That being said, if you have time, increasing the count will always make your data more accurate.

Remember that the test looks for a correlation between all variables of data, if one or several variables have no relation, then the power of the test will go down.

Now, if you want to increase your statistical power, what you need to do is actually remove variables that are independent from the rest. Example, certain factors are more closely related then other factors. After running your test you should see how each affects the overall result. If any are particularly weak, remove them (one at a time) and run the test again. Using your 10 strongest (or whatever # you choose) variables might increase your statistical strength of your test.

 

edit: simple example:

You are trying to determine someone's income based on education, house size, and eye color. Running the test you'd probably expect education (higher education, more money) and house size (larger house, more money) to be the primary factors of someone's income. While eye color is probably independent of the other variables. Running the regression test with all three variables anyway will pull down the significance of the test, while removing eye color and running the test again with just education and house size will (likely) increase it. Though if you discover blue eyes give an advantage, let me know :)

While a bit extreme in the above example, the princible holds the same. There might be some variables pulling the weight of your test down.

Related Questions

Navigate

• Browse (All)
• Browse H
• Subjects

---

---

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.