|
baphd1996
|
 |
« on: November 05, 2009, 11:06:11 AM » |
|
It has been a long time since I've had to use statistics, so I'm a little rusty, but back in the day...
About a year ago I made several changes to my course. I increased the number of exams and quizzes, I required the students to draw structures and turn those drawings in for a grade, and I became hardnose about spelling (if it was spelled wrong, the answer was wrong). Now I want to show my bosses that my changes made a difference.
I have 9 sets of grades from classes before I made the change and and 5 sets of grades from after the change. My plan is to compare the final percentages for the before and after groups.
So, should I treat each class as separate samples, or should I combine all of the before groups into one and all of the after groups into one? What statistic should I use? Is this even a valid comparison due to all of the variables (different amounts of possible points, different sample sizes, differences due to counting off for spelling)?
Any ideas?
Thanks for your help[!
|
|
|
|
|
Logged
|
I don't have time to read what I wrote!
|
|
|
|
august_leo
|
|
« Reply #1 on: November 05, 2009, 12:02:13 PM » |
|
Try:
1. Take the percentage for every student before (pre group) and the percentage for every student after (post group) and conduct an unpaired t-test to see if they are significantly different. You should have 1 score per student, so if you had 10 students in each class, you would be comparing 90 scores to 50 scores. If you are using SPSS or something similar, you will have a column of percentages and a column of group (pre, post) and the same number of rows as total students.
2. Alternatively, do the above, but take the 5 most recent classes before (pre group) an compare them to the 5 post groups.
3. You can also enter the scores for each student, the group (pre, post) and a third column of class (one, two,... fourteen). Then you can do an ANOVA or MANOVA with class as an independent variable in addition to group.
|
|
|
|
|
Logged
|
Your environment sounds vaguely toxic. Or maybe just characteristically British.
I heart august_leo.
|
|
|
|
namazu
|
|
« Reply #2 on: November 05, 2009, 12:10:00 PM » |
|
A graphical, non-hypothesis-testing way to display the data would be to do box-plots for final scores for each class, in chronological order. So you'd have 14 box-plots side by side, and could draw a line between them where you instituted the more rigorous requirements/expectations. This would give an at-a-glance view of where the median and quartile score fell before and after the change, and what kind of grade range there was within classes.
|
|
|
|
|
Logged
|
|
|
|
|
offthemarket
|
|
« Reply #3 on: November 05, 2009, 12:14:51 PM » |
|
Do a two-way ANOVA, with one categorical variable being class section, and the other being before/after change. If there is no interaction effect between the two variables, then the change in grades is the result of the change itself, not because of idiosyncracies related to class section.
I think.
|
|
|
|
|
Logged
|
|
|
|
|
baphd1996
|
|
« Reply #4 on: November 05, 2009, 12:51:10 PM » |
|
Thank You for all of your suggestions. I had already turned the scores into percentages to "normalize" the grades, so I'm glad I'm on the right track. I was unsure about the T-test or ANOVA so thanks.
|
|
|
|
|
Logged
|
I don't have time to read what I wrote!
|
|
|
|
mad_doctor
|
|
« Reply #5 on: November 05, 2009, 01:34:55 PM » |
|
Use an HLM model, set the covariance matrix to auto-correlative, and test to see if there are between-class differences (use AIC, if I recall correctly). You may have to do both tests separately, depending on your software. If either of the tests is positive, then there is a temporal or class effect, or both depending on which test is positive. If the tests for auto-correlation and between-class are negative, then you can group them all together and do between-treatment comparisons.
|
|
|
|
|
Logged
|
|
|
|
|
