Information within residency monthly evaluation forms at two institutions

Figures & data

Table 1. Basic information from Marshall and University of Toledo Programs.

	Marshall University		University of Toledo
Program	Evaluations(N)	Likert Questions (N)	Evaluations (N)	Likert Questions (N)
Psychiatry	84	65	67	33
Surgery	491	7	642	15
Internal Medicine	728	29	520	26
Pediatrics	453	22	344	9
Obstetrics	742	16	109	22*
Family Medicine	372	9	61	30

University of Toledo Obstetrics used a number of different evaluation forms for different rotations and evaluators. Correlation analysis was not felt possible with the limited sample size for each form.

Figure 1. Panel A: Histogram of resident scores in Marshall IM for question 3. In contrast for the third subject in the Marshall IM program, all Likert scores were exactly 4. Panel B: Boxplot of Average Variance in Likert Score within each resident. Panel C: Boxplot of Average Variance in Likert Score across questions amongst (between) all residents. Data in Panels B and C shown as mean ± SD. Note that with one exception (UT Pediatrics), the average variance in Likert scores is greater between residents than within residents.

Figure 2. Box plots showing distribution of scores per individual resident. Median marked with solid line, box traverses 1^st and 3^rd quartile. Outlier values marked as solid circles. Data demonstrates the variability of scoring is greater between residents (residents numbered on x axis) than between Likert questions applied to the individual resident (y range) in somewhat different way that Figure 1 above.

Table 2. Correlations on forms from Marshall and University of Toledo Programs.

	Marshall University		University of Toledo
Program	Mean	SD	Mean	SD
Psychiatry	0.88	0.04	0.45	0.34
Surgery	0.91	0.05	0.60	0.30
Internal Medicine	0.91	0.03	0.66	0.22
Pediatrics	0.83	0.08	0.46	0.37
Obstetrics	0.84	0.07
Family Medicine	0.74	0.12	0.59	0.16

Figure 3. Correlation heat maps with network groupings shown for Marshall (top Panel) and UT (bottom Panel) residency programs. Color code for correlations shown on graphs. Network groupings fixed by R software using the command heatmap.2 within the package gplots. The results depict correlation between items with highly correlated items coded light green, as demonstrated by the line of identity for identical items that are displayed from bottom left to top right, to less well correlated items that are depicted in in darker shades.

Figure 4. Principal component analysis of Marshall Residency programs. Note that for all programs, the first principal component contained the vast majority of information. In fact, for all programs, more than 95% of variance was contained within first 5 principal components. For FM, the distribution of significant variance went up to the 5^th component whereas for other programs, it was limited to the first 1 or 2.