Intro

Below, I am going to show you how I would go about computing the stats for the Simulation project. I am going to use the “UCSF Student Video RAW data” (that I pre-cleaned) as the practice data. Don’t pay much attention to the code, I’ll try and narrate the process as I go along.

Data Import

## Source: local data frame [6 x 4]
## 
##     School CLecPrefVC CTextPrefVC COnRecLecPrefVC
## 1  Nursing          4           4               5
## 2 Medicine          4           5               5
## 3 Medicine          4           2               3
## 4  Nursing          4           5               4
## 5 Medicine          4           4               4
## 6 Pharmacy          4           4               3

Above is the first 6 entries from the dataset. You’ll notice that I altered the text responses to follow a 1-5 likeart scale. Next, I’ll make a couple basic graphs to have a look at the data. (Note that, in this instance, it would inappropriate to graph each group in the same plot due to the groups containing unequal subjects.)

Graphs

Obviously, these data don’t follow a normal gaussian distribution, so a Kruskal-Wallis test will be the most appropriate.

Statistical Analyses

First, lets have a look at the results for the “Compared to Lectures, I prefer Video Cases” set.

## 
##  Kruskal-Wallis rank sum test
## 
## data:  CLecPrefVC by School
## Kruskal-Wallis chi-squared = 8.3817, df = 2, p-value = 0.01513

The results of this test suggests that there is a significant difference between the groups (p=0.0153). However, the limitation of this test (as with standard ANOVA), is it does not tell you where the difference is, or which group preferred what the most (or least). All that we know for sure is that, if there was truly no difference between these groups the probability of finding results at least as similar to this is about 2/100. We might be able to take an educated guess as to which group preferred this the most (in this case, the “pharmacy” group), however it’s just a guess.

Let’s now take a look at the “Compared to Online Recorded Lecture, I prefer video cases” responses. (I skipped the “Compared to Textbooks, I prefer Video Cases” responses because it, too, shows a significant difference.)

## 
##  Kruskal-Wallis rank sum test
## 
## data:  COnRecLecPrefVC by School
## Kruskal-Wallis chi-squared = 1.7937, df = 2, p-value = 0.4079

These results suggest that there is no significant difference between the groups (p=0.4079).

Optional Graphing Technique for Unequal Groups

Because the subgroups we are comparing have an unequal number of subjects, we could graph this by proportion instead of count. If we did this, here’s what the graphs would look like…

This is still imperfect, but it’s another way of looking at it.

Let me know your thoughts! :)