387038 Does Time-to-Degree Matter? Survival Analysis and Students' Scores Evaluation

Wednesday, November 19, 2014: 3:20 PM
M106 - M107 (Marriott Marquis Atlanta)
Adelmo Filho, Karla Esquerre and Elaine Albuquerque, Chemical Engineering, Federal University of Bahia, Salvador, Brazil

Does time-to-degree matter? Survival analysis and students’ scores evaluation

The undergraduation dropout and the long duration are remarkable features of the current teaching of Chemical Engineering in Brazil. In general, the course proposes the formation of a multidisciplinary professional, capable of acting in the management, operation and control of chemical and petrochemical industries. The Chemical Engineering Undergraduation Course at Federal University of Bahia is over 50 years old and is among the 10 state colleges with major prestige in the labor market in Brazil. The last few years have been particularly important for the course because of the change of the curriculum in 2008 and the change of the approval average from 7 to 5 on a scale of 0 to 10. In this context, this research attempts to make a statistical analysis of the students’ scores from 1971 to the present with a focus on modeling the time of conclusion or abandonment from the course through the survival analysis.

Questions like, if there were changes in the length of time needed to complete the Chemical Engineering or on their score over time or whether it would be possible to predict the likelihood of the graduation at the present time based on the students’  score and semester sought be understood. In the analysis of survival, the time until an event of interest is adjusted by means of asymmetric probability curves with the possibility of using censored data (length of stay of students who have not yet completed or dropped out of the course). Based on the interpretation of the shape of the distribution and the value of its parameters is possible to obtain important information about the phenomenon of interest. The statistical analysis was developed with graphs of correlation between the variables (year of entry, score and time of conclusion/ dropout of the course), boxplot charts and histograms to verify initial assumptions that are important for survival analysis and to verify the hypothesis of randomness of the variables along the time. The entire study was conducted through open language R through the R-Studio GUI with the function package called Survival. The database had approximately 3300 students' data classified by the way they entered and left the college. There are a few outliers or inconsistent information, which was eliminated in the process. The behavior of the Chemical Engineering students' score has shown to be superior to the consensus with median and average values equal to 6.9 and 6.7, respectively, and with right asymmetry distribution. The application of the Fisher test for non-randomness in time series rejected the null hypothesis against a p-value less than 0.05, which indicates a periodic behavior. When the same test was applied to the time series of the score, the results were similar, indicating the frequency of this variable over time. The correlation between the score and time-to-degree shows a triangular shape. These variables are negatively correlated when the students’ score are over five and are positively correlated when students’ score are below five.

Survival analysis was conducted with two groups of data: (1) students who graduated and (2) students who dropout the course. The data sets were modeled by using exponential, Weibull, log-normal, Gaussian, log-logistic and logistic distributions. Kaplan-Meier estimator was used to compare the parametric distribution with the non-parametric survival curve, and to select the best model.. The first analysis indicated a significant fit of the data with the Weibull distribution (scale = 1.16 and shape = 0.0688). The shape's value indicates a distribution with “Child mortality” that means in our terms a well-expressed capacity of the students in finish the course on less time. The median was equal five, value that correspond to the time expected by the course administration for the course conclusion. The second group of data obtained best fit with the exponential distribution (scale = 0.937). The histogram indicated that the greater number of dropouts happens during 3rd and 4th semesters of the course. At this stage of the course, the students start to deal with chemical engineering aspects in chemical process on industry, statistics to engineers and computational methods for engineering courses. The other courses are related with theoretical math and physics.

It is expected that the obtained results contribute to uncover the student’s profiles involving the student’s time-to-degree and scores, and base the development of long term enrollment plans at the Polytechnic School and central campus levels.

Extended Abstract: File Not Uploaded
See more of this Session: Free Forum on Engineering Education: Junior and Senior Years
See more of this Group/Topical: Education Division