Descriptive results: Figure 1 depicts the difference in attrition rate between males and females, per specialty. This gender difference (projected on the x-axis) is positive if the attrition rate is higher for males in training, and negative if the difference is higher for females. Specialties are sorted on the y-axis by the proportion of females in training between 2003 and 2013.
Outlier analysis: In figure 1, the specialty cardiothoracic surgery is excluded as this appears to a clear outlier. For this specialty the difference in attrition between males and females was ─42% (i.e. the attrition rate for female trainees in cardiothoracic surgery was 42% higher compared to male trainees), where the average for all specialties was ─2,7%. This outlier result is probably due to the fact that the attrition rate for cardiothoracic surgery is based on a total of only 62 trainees in 10 years (2003-2013), while the average number of trainees for the other training programs is 445. Including this outlier clearly skewed the distribution of the dependent variable Y1 in the data set (Anderson-Darling value: 1.260), whereas excluding it changed the distribution to a normal distribution (Anderson-Darling value: 0.886). The outlier analysis is depicted by Figures 2 (with cardiothoracic surgery) and 3 (without cardiothoracic surgery). Based on this, it was decided to exclude cardiothoracic surgery as a specialty for further analysis.
Next, four linear regression analyses were executed to explain the variation in Y1, gender differences in attrition rate, by the dependent variables X1 to X4 at the level of the 25 specialties or training programmes (hence excluding cardiothoracic surgery). Below we present the results in order of the prediction strength of the dependent variables. After this, and as mentioned above, a Kruskal-Wallis test was performed to analyse the relation between the gender differences in attrition rate and type specialty training (X5).
Results of the regression analyses showed the strongest predictor for X2, the proportion of males working in the profession per 01-01-2003. More than a half (55%) of the variation in the difference in attrition rates is explained by this dependent variable, a significant finding (R2: 0.545, F(27.60), p<.000). Figure 4 depicts this result. The regression model shows that residuals were normally distributed, giving no signs for other outliers. Also, the prediction interval for the regression model equation felt within the 95% boundaries.
The negative and significant coefficient implies that the lower the percentage of males working in a speciality, the higher the difference in attrition rate between males and females that are training for that specialty (top left hand of the fitted line plot, figure 4). Specifically, in the specialties with lower percentages of males working, males have higher drop-out rates compared to females in training. The opposite can obviously be concluded as well, as the lower right end of the fitted line plot in figure 4 implies: in specialties with more males working in that profession, the drop-out rate of females in training is significantly higher.
Another regression analysis was done for the relationship between Y1 and the X1, the proportion of males in training. This analysis showed that 42% of the variation in the gender difference in attrition rates was explained by this predictor, as significant finding (R2: 0.417, F(16.01), p<0.000), and depicted by Figure 5. Residuals were normally distributed, the prediction interval for this regression equation felt within the 95% boundaries. This negative and significant coefficient implies that the lower the percentages of males in training in a specialty, the higher the difference in attrition between males and females in training for that speciality (top left hand of the fitted line plot, figure 5). In contrast to the previous regression analysis result however, the specific result is that less male residents drop-out if the proportion of males is higher in a specialty. And vice versa, shown by the lower right end of the fitted line plot (figure 5): if specialties have more males in training, more females will stop their training before completion, i.e. females have significant higher attrition rates.
A third regression analysis was done to analyse the relationship between difference in male and female attrition rates and the total attrition rate for each training program (X3). The result can be summarized as a relative small but significant effect (R2: 0.163, F(4.46), p<0.046). From Figure 6, it can be derived that for specialty training programs where the overall attrition rates are higher, the difference in attrition between males and females is higher as well (top/right side of figure 6). Likewise, in specialties where the overall drop-out rates in the training program are lower, the difference in attrition between males and females in training for that speciality is lower as well (bottom/left side of figure 6).
A fourth regression analysis was performed to explore the relationship between the difference in attrition between males and females in training, and the duration of the specialty training (X4). In sum, it can be showed this relation is negative and significant (R2: 0.299, F(9.85), p<0.005), see figure 7. Longer specialty training programs thus show larger differences in attrition between males and females in training.
In a final step, the differences between the specialties as such were explored (X5), in particular the distinction between surgical and other specialties. This is partly related to the previous analysis as most surgical specialties also have the highest training duration. A Kruskal-Wallis test was preformed to analyse the difference in attrition between males and females in training by type of specialty. The difference between the types of specialisms were significant (H=6.66, p=0.036). The ranking results of the Kruskal-Wallis test implied that the gender differences in attrition rate was negative for ‘surgical’ specialisms (Z-value -1.98), positive for ‘auxillary’ specialisms (Z-value 2.29) and ranked in between for the ‘non-surgical’ specialisms (Z-value -0.11).