IBM Watson Analytics TRENDS FROM STUDENT RETENTION DATA SET ANTHONY VALLONE MIS3063-601 DR. BRENDA KILLINGSWORTH Overview The purpose of this report is to detail findings discovered by exploring a student retention data set using IBM Watson Analytics. Topics included in this report include the trends associated with a student's high school GPA and their performance in other areas, including graduating. Also, this report explores patterns and findings broken down across ethnic lines, and explains these patterns further. This report also discusses the top indicators of whether or not a student who attends university will graduate, and settles the age old question: who rules and who drools. The Data The data used in this report is a student retention excel spreadsheet that tracks information related to students who have been admitted to a University. Several categories are included in the data: gender, ethnicity, birthdate, high school attended, high school GPA, math placement score, algebra placement score, reading placement score, writing placement score, full/part time, male/female guardian education level, category of major, whether or not the student attended orientation, and whether or not the student graduated. The data was obtained from six high schools, and only four ethnic groups - White, Black, Hispanic, and Asian. In total, 1550 students were recorded in this student retention data set. Using this data set, further exploration into trends within these student retention data was made possible Exploring Trends in High School GPA Everyone can relate to having a high school GPA, and that is why I chose to explore this metric first. High school GPA, it turns out, can tell us many things about how well a student will excel in a university setting. Several illustrations are presented below relating to findings obtained by exploring a student's high school GPA, and this report will further explain what these illustrations tell us: Page | 1 Figure 1 The graphical display in the upper-left of figure 1 is a heat map that shows the concentration of students that did and did not graduate, based on their GPA. As you move from left to right across the heat map, the high school GPA is going upward - starting from 0 to 1.15 in the first column, and going all the way to 4.68 to 5.85 in the last column. The graph is then further divided in by two rows. The top row shows the concentration of students that did not graduate, and the bottom row shows those that did. The visualization shows that students that had a higher high school GPA tended to graduate much more of the time then those who had low GPAs. From looking at the graph in the upper-right of figure 1, it becomes apparent that students with Humanity majors tended to have much lower high school GPAs then those of other major; furthermore, the Math/Science students seemed to have exceled in high school. Given that GPA correlates with whether or not a student tended to graduate, I wanted to understand what contributed most to a student's GPA. The driver's list presented in the lower-left of figure 1 shows that the leading factors that contributed to a student's GPA was the education level of his/her parents. The graph in the lower-right of figure 1 illustrates this point. As the education level of the parents increased from less than high school to a 4 year degree and up, there is a significant increase in the average GPA of their children. There may be several explanations for this finding - such as the parents being able to assist their children, or fostering a positive environment for learning - but the trend is definitely there. In order to further see the inclination of the different majors and the relationship to a student's GPA, I created a line graph that shows the average GPA of students in various majors: Page | 2 Figure 2 The line graph spans from 1989 to 1995 in order to illustrate the trends in the average GPA of students with different majors. This graph makes it obvious that, on average, Humanity majors, have significantly lower GPAs than those in other majors. Additionally, we see that the average GPAs have gone up over this 5 year period. Exploring Ethnic Trends After looking at some developments relating to GPA, I wanted to break the data along ethnic lines to further explore some patterns in the student retention data. Recently, there have been discussions on institutional biases against persons of color, and I wanted to see if that was the Page | 3 case with this university: Figure 3 The first thing that we can note from figure 3 is that the majority of students admitted to this university, from the six high schools in the data set, are Hispanic. The left portion of figure 3 is broken down by the four ethnic races in the data, and also on whether or not these students graduated. A positive sign that we glean from the visualization, there is no signs of racial favoritism in terms of the graduation rate of each race. In fact, the graduation rate across all four races were within 5 percentage points (on average 75% graduated). The right portion of figure 3 shows the average GPA of students from the four ethnic groups. There seems to be relationship between race and high school GPA, so I dug deeper to determine why this is the case: Page | 4 Figure 4 Breaking the student retention data set down further, I discovered that the average GPAs correlated more with the high school the students attended more so than ethnicity. On the left, figure 4 shows the number of students that attended each of the six high schools and is broken down by ethnicity. On the right, figure 4 shows the average reading score (y axis), average math score (x axis), and the average GPA is illustrated by the size of the circle on the graph - the bigger the circle, the larger the average GPA. It turns out that Blacks and Hispanics, on average, attended the poorer performing high schools - accounting for the ethnic differences in GPA. The biggest trend that broke down along ethnic lines is shown below: Figure 5 Page | 5 Figure 5, above, is a graph the shows the number of students that attended the university's orientation broken down along the four ethnicities. Across almost all ethnicities, students were more likely to not attend orientation than to go to the orientation. However, this is absolutely not true of Asians. Asians are four times more likely to go to the university's orientation than not go. This was by far and away the biggest difference across all metrics, in terms of race. Exploring Trends in Graduating Ultimately, the reason students go to university is to graduate and earn a degree. For this reason, I consider the following section the most valuable of the report. While examining high school GPA and ethnic patterns may be interesting to some, everyone who attends university can appreciate the patterns and trends of students who graduate. To understand the biggest influences on whether or not a student graduated, I broke the data into factors that correlated most with graduation - illustrated below: Figure 6 Figure 6 breaks down whether or not a student graduated by categories that correlated most with successful graduation. The number 1 and 2 predictors of whether or not a student would graduate was the education level of the student's male and female guardians, respectively. This point is further explored in the graph below: Page | 6 Figure 7 The left side of figure 7 shows the female guardian education level, while the right side shows the male guardian education level. The y axis illustrates the percentage of students that did graduate (the upper portion of the bars) and the percentage of students that did not (lower portion). The percentage of students whose male or female guardian had less than a high school degree, were more than 30% likely to not graduate. Juxtaposed to this, students whose male or female guardian had a 4 year degree or higher, were only 10% likely to not graduate. In fact, as the education level of the guardian increased, so did the likelihood of the student to graduate. Though the education level of the guardians were the top predictors of student success, there are some other notable ones: Page | 7 Figure 8 The upper-left portion of figure 8 shows that the majors that students were in was had trends of whether or not they would graduate. The Humanity majors were lease likely to graduate (with a graduation rate of only 60%), while the Engineering majors were the most likely to graduate (88%). The upper-right section shows the number of students that did or did not graduate based on whether or not they attended the university full or part time. Full time students were 5 times more likely to graduate than not, compared to part time students who were barely favored to graduate than not. The lower-left portion of figure 8 details the relationship between academic performances - as measured by the metrics: reading placement scores, math placement scores, and high school GPA - and graduation rate. As you may expect, the students that excelled academically were significantly more likely to graduate than those who performed poorly in these areas. The lowerleft of figure 8 was discussed earlier in figure 1, but it is worth mentioning again that the students with higher high school GPAs graduated at higher rates. Battle of the Sexes What would a comparison of performance be without a competition between boys and girls? I felt obligated to break the data down into performance metrics and pit one side against the other to settle once and for all, who rules and who drools. The results are illustrated below in figure 9: Page | 8 Figure 9 I broke the battle of the sexes down into 4 performance metrics: 1. 2. 3. 4. Average high school GPA (upper-left): girls 3.42, boys 3.20 Average reading placement score (upper-right): girls 336.50, boys 350.78 Average math placement score (lower-left): girls 330.78, boys 350.65 Percentage of those who graduated (lower-right): girls 78%, boys 73% Sorry boys, the numbers just do not lie. The girls performed better than the boys in every metric tested. In Conclusion Using IBM Watson to explore a student retention data set revealed and confirmed a number of things. For one, the indicators of success in university are high school GPA, education level of guardians, choice of majors, and placement scores. Furthermore, based off of the data explored here we can say that there is zero indication based on ethnicity, whether or not a student will graduate from this university and debunks any notion that this university participates in institutional biases against people of color. Lastly, at least according to this data set, female students out preformed boys in every metric - including the percentage of those that graduated. Page | 9 References IBM, IBM Watson Analytics. Available at: http://www.ibm.com/analytics/watson-analytics/ [Accessed July 11, 2017]. Killingsworth, B. (2017, June 20). Student Retention Data(1). In ECU Blackboard Learn. Retrieved July 10, 2017, from https://blackboard.ecu.edu/webapps/blackboard/content/listContent.jsp? course_id=_427774_1&content_id=_9228072_1&mode=reset Page | 10