Planning & Research Office Report 20150185 Requestor: Math Department Researcher(s): Steve Blohm and Terrence Willett Date: 6/30/15 Title: Effects of Math Tutoring Effects of Math Tutoring Introduction The purpose of this study is to measure the effects of math tutoring at Cabrillo College. This is an observational study in which we looked at student grades, success rates, and completion rates for students who received tutoring and those who did not in all math classes in which at least one student was tutored. From a research point of view, it would have been better to use an experiment where we could examine the effects of tutoring while holding other variables constant. In order to perform an experiment to measure the effect of math tutoring, we would need to randomly select students to be tutored from a pool of students who seek tutoring, while denying other students tutoring. However, due to practical, ethical, and regulatory concerns about denying tutoring students to students, an experimental approach was not possible. The direct comparisons between those receiving tutoring and those not receiving tutoring while useful are subject to self-selection bias. To help control for bias from students selfselecting to participate in tutoring, regression and propensity score matching (PSM) techniques were employed to equate participants and non-participants on a variety of background variables. The results are of this study are consistent with the belief that tutoring is helpful for students. Sample The sample analyzed consists of students who used tutoring at the Math Learning Center (MLC) from summer 2009 through summer 2013 along with those who were enrolled in a course where at least one of the students used tutoring. The MLC is overseen by a math faculty member and provides free, drop in math tutoring for Cabrillo students in a study hall setting including closed study rooms and computer stations1. Math tutors go through a training class with regular workshops to provide general tutoring skills and to ensure students with learning disabilities and other special needs can be accommodated. The MLC allows students to check out math textbooks, graphing and scientific calculators, laptops, 1 https://www.cabrillo.edu/services/mlc/ 1 Planning & Research Office Report 20150185 software, audio visual materials, and “learning manipulatives” such as blocks to aid understanding of fractions and proportions. We looked at six courses at Cabrillo where enrollment and tutoring use were both relatively high compared to other courses. The courses we looked at were Statistics (Math 12), Intermediate Algebra (Math 152), Basic Algebra (Math 154), Essential Mathematics (Math 254A), Pre-calculus (Math 4), and Calculus (Math 5A). Summary Findings Many students at Cabrillo have had difficulty succeeding in Intermediate Algebra. We look in detail at the relationship between hours tutored and success in this course while controlling for other factors that are predictive of success. We came up with an equation that indicates the number of tutoring hours that are required on average for a particular probability of success for a given course given certain demographic indicators. While our study indicated that students in general will have a greater chance of success if they receive tutoring, using this equation we can come up with some concrete numbers. For example, to have at least a 75% chance of success in intermediate algebra, an underrepresented minority (URM) male would need, on average, 29.5 hours of tutoring in a semester or about two hours of tutoring per week. A URM female would need 23.24 hours of tutoring for the same probability of success. Detailed Findings The following figure looks at the success rates in each class for these three groups 1. Students who did not log tutoring time or log time at the MLC - Nothing 2. Students who used the MLC, but did not use tutoring – MLC Only 3. Student who used some tutoring – Tutoring 2 Planning & Research Office Report 20150185 Figure 1: Percentage of Math Students Earning a C or Better In every case, those who were tutored outperform those who were not. The sample sizes were large so even small difference would show up as significant. However, in many cases the differences are substantial. For example, pre-calculus students who were tutored succeed 58% of the time and those who were not succeeded 48% of the time. Below we will investigate the effect of tutoring among specific among different gender, ethnic groups, and other special populations. 3 Planning & Research Office Report 20150185 Figure 2: Percentage of Math Students Earning a C or Better by Gender Both female and male students perform better with tutoring than without. Female students tend to outperform male students in all cases except for those not tutored in Calculus I. In every case other than Math 154, Elementary Algebra, the higher success rate for female tutored students was statistically significant at the 0.05 level of significance. For male students the higher success rate for tutored students was statistically significant in all cases other than Math 12 (Statistics) and Math 5A (Calculus). Chi-Square details are in Appendix A. 4 Planning & Research Office Report 20150185 Figure 3: Percentage of Math Students Earning a C or Better by URM Status While under-represented minority students (URM) do tend to have lower success rates compared to other students, those URM students who receive tutoring perform better than URM students who do not receive tutoring. For Math 152, Math 254A and Math 5A the higher success rates are statistically significant at the 0.05 level significance. This information is based on a two-tailed significance level where the hypothesis is whether or not success rates are different for tutored students. If we were to instead consider these one-tailed tests where our hypothesis is whether or not tutored students have a higher success rate then the results would be significant for all six courses2. Chi-Square details are in Appendix A. 2 While a chi-square test is generally only a two-tailed test because the chi-square statistic is not symmetrical, in this case our chi-square is mathematically equivalent to a Z-statistic. We can therefor take half the p-value of the two tailed test and compare that to the 0.05 level of significance for a one-tailed test. 5 Planning & Research Office Report 20150185 Percent of Students Enrolled in Intermediate Algebra that Used Tutoring by Ethnicity 29.82% 25.63% 23.68% 23.34% 18.68% 24.14% 18.52% 13.45% 8.11% American Indian, Alaskan Nativ Asian Black NonHispanic Filipino Latino Multiple Ethnicities Pacific Islander Unknown White NonHispanic Figure 4: Tutoring Use By Ethnicity Of the 8,082 Students enrolled in math 152 over the years studied, 1,757 (21.74%) used tutoring. 19.25% of under-represented minority (URM) students used tutoring for intermediate algebra while 23.95% of non-URM students used tutoring. The table below shows tutoring use by ethnicity for Math 152, Intermediate Algebra. While URM students in general used less tutoring than other students, black students used tutoring at the highest rate. 6 Planning & Research Office Report 20150185 Figure 5: propensity score matching analysis of tutoring use The above chart shows performance differences between students who were tutored and a comparison group of students who were not tutored but have similar attributes based on one to many propensity score matching (PSM). The criteria used for matching were gender, ethnicity (as a binary variable indicating URM or not), age, and prior overall grade point average (GPA). The chart shows that on average those who were tutored had better success rates than those who were not. The chart also shows a 95% confidence interval for the difference. If the confidence interval does not include zero then we can be at least 95% confident that the difference did not occur by random chance. In every case except for Math 154, Elementary Algebra the results were statistically significant. The largest difference in success rates between the two groups ids for Math 254, Pre-algebra. 7 Planning & Research Office Report 20150185 Figure 6: Distribution of Hours Tutored for Math 152 Intermediate Algebra For Math 152, Intermediate Algebra, the maximum number of hours someone was tutored for a semester was 72 hours and the minimum was less than an hour. The mean time for tutoring was about 3 hours per semester with a standard deviation of 5.5 hours. But as can be seen in the histogram above, the data are not normally distributed. The median for hours tutored was 1.15 hours; 50% of students tutored received 1.15 or less hours of tutoring in the semester. Seventy five percent of students tutored received 3.16 or less hours of tutoring during the semester. Looking at these numbers separately for those who succeeded vs those who did not succeed we see the average number of tutoring hours for those who succeed was 3.58 and the average number of tutoring hours for those who do not succeed was 2.29. This difference is statistically significant at p < 0.001 8 Planning & Research Office Report 20150185 Logistic Regression We used a Logistic Regression to estimate the number of tutoring hours required to succeed in Intermediate algebra and to attempt to at least partly control for self-selection bias. Below we estimate the ln(odds of succeeding) using hours tutored, gender, URM status, tutoring use and overall GPA . π = ππππππππππ‘π¦ ππ π π’ππππππππ ππ πππ‘β 152 π ππ ( ) = πΌ + π½1 π₯1 + π½2 π₯2 + π½3 π₯3 + π½4 π₯4 + π½5 π₯5 + π 1−π π ππ ( ) = πΌ + ππ’π‘ππππππ»ππ’ππ π₯1 + ππ ππ₯2 + πΊππππππ₯3 + πΊππ΄π₯4 + ππ πππ’π‘πππ₯5 + π 1−π Variables in the Equation Constant Tutoring Hours URM (Y = 1) Gender (F = 1) Overall GPA Use Tutoring (Y = 1) Beta -0.541 0.046 -0.420 0.339 0.229 0.104 Std. Error 0.055 0.012 0.046 0.046 0.017 0.063 Wald 95.802 16.288 83.445 54.537 172.508 2.668 df 1 1 1 1 1 1 P-Value P < 0.001 P < 0.001 P < 0.001 P < 0.001 P < 0.001 P = 0.102 Because we are looking at the log odds of succeeding, it is difficult to interpret the Betas. However, positive values mean that the variable has a positive influence on succeeding and negative values mean the variable has a negative impact on succeeding. Of the variables coded dichotomously, URM status has the largest influence. We could use the equation below to predict the probability of success for a particular student that student’s information for each of the variable in the equation. ππ ( π ) = − 0.541 + 0.046π₯1 − 0.420π₯2 + 0.339π₯3 + 0.229π₯4 + 0.104π₯5 + π 1−π 9 Planning & Research Office Report 20150185 Alternatively we could use the equation to estimate the Number of Tutoring Hours Required to Succeed in Intermediate Algebra The equation below can be used to predict the ln(odds of succeeding) using hours tutored, gender, URM status and overall GPA as a covariate. π = ππππππππππ‘π¦ ππ π π’ππππππππ ππ πππ‘β 152 π ππ ( ) = πΌ + π½1 π₯1 + π½2 π₯2 + π½3 π₯3 + π½4 π₯4 + π 1−π π ππ ( ) = πΌ + ππ’π‘ππππππ»ππ’ππ π₯1 + ππ ππ₯2 + πΊππππππ₯3 + πΊππ΄π₯4 + π 1−π Variables in the Equation Constant Tutoring Hours URM (Y = 1) Gender (F = 1) Overall GPA ππ ( Beta -0.142 0.048 -0.610 0.303 0.148 Std. Error 0.133 0.012 0.100 0.100 0.040 Wald 1.15 17.39 37.15 9.21 13.43 df 1 1 1 1 1 P-Value P = 0.284 P < 0.001 P < 0.001 P = 0.002 P < 0.001 π ) = − 0.142 + 0.048π₯1 − 0.610π₯2 + 0.303π₯3 + 0.148π₯4 + π 1−π Note: Sample only includes those who used tutoring. 10 Planning & Research Office Report 20150185 If we want to estimate the number of tutoring hours required for success for a group, it is simpler to exclude GPA (a continuous variable) as a covariate. Here we estimate the ln(odds of succeeding) using hours tutored, gender, and URM status. π = ππππππππππ‘π¦ ππ π π’ππππππππ ππ πππ‘β 152 π ππ ( ) = πΌ + π½1 π₯1 + π½2 π₯2 + π½3 π₯3 + π 1−π π ππ ( ) = πΌ + ππ’π‘ππππππ»ππ’ππ π₯1 + ππ ππ₯2 + πΊππππππ₯3 + π 1−π Variables in the Equation Constant Tutoring Hours URM (Y = 1) Gender (F = 1) ππ ( Beta 0.213 0.052 -0.645 0.322 Std. Error 0.091 0.012 0.099 0.099 Wald 5.499 20.32 42.169 10.52 df 1 1 1 1 P-Value p = 0.019 p < 0.001 p < 0.001 p = 0.001 π ) = 0.213 + 0.052π₯1 − 0.645π₯2 + 0.322π₯3 + π 1−π Note: Sample only includes those who used tutoring. 11 Planning & Research Office Report 20150185 Now we can estimate the number of tutoring hours required for success in Math 152 for a particular population of interest. For example, if we want to know the number of hours of tutoring estimated for under-represented minority male success in Intermediate Algebra we could use the equation below. The equation for an under-represented minority male simplifies to: π ππ ( ) = − 0.432 + 0.052π₯1 + π 1−π • • • • • • Among the 1,467 Male URM students who were not tutored, 560 succeeded (38%) Among the 316 Male URM students who were tutored at all, 129 succeeded (41%) For p = 0.5 we estimate that a URM male needs 8.3 hours of tutoring (about 0.5 hours per week) 29 Male URM students were tutored for at least 8.3 hours and 17 succeeded (59%) For p = 0.75 we estimate that a URM male needs 29.5 hours of tutoring (about 2 hours per week) The two URM male students who had at least 29.5 hours of tutoring both succeeded. 12 Planning & Research Office Report 20150185 English Placement Scores Other covariates such as English Placement scores could be used. However, comparing English Placement scores for those who used tutoring and those who do not use tutoring we see no significant difference. Used Tutoring N No Yes 5,134 1,271 Mean Std. English Deviation Placement Score 45.85 8.143 45.48 8.183 Std. Error Mean 0.114 0.230 The difference of 0.37 is not significant (p = 0.149). Given that the sample size is over 6,000, this is evidence that English placement scores are not related to the decision to use tutoring for Math 152, Intermediate Algebra. There is a significant difference between English placement scores for those who succeed in Math 152, Intermediate Algebra vs those who do not succeed in Math 152; English placement is a significant predictor of success in Math 152. Success in Math152 N No Yes 3,314 3,091 Mean Std. English Deviation Placement Score 44.57 8.111 47.07 7.996 Std. Error Mean 0.141 0.144 The difference of 2.5 is statistically significant (p < 0.001) When we add English placement scores to the equation it is a significant predictor of success. However, the fit of the equation does not change in any practical way and the sample size is reduced by almost 1,700 students when we use this information. The primary implication is that more tutoring did appear to lead to greater success in math classes in general. Even those students who use tutoring tend to use very little. Encouraging students to use tutoring and to use it often would likely lead to greater success in math courses. Some of the underrepresented minority groups used tutoring at a disproportionately lower rate than their white counterparts. In particular, Latino, Filipino, and Pacific Islander students used tutoring at less than 80% of the rate of use by White students. Other non-URM Asian students tended to use less tutoring as well. No more than 30% of any ethnic group used tutoring and this speaks to the fact that all students, regardless of their background, could probably benefit by using more tutoring. 13 Planning & Research Office Report 20150185 Appendix A Female Students Course Name Chi-Square N df Sig. (2-sided) MATH-12 14.661 2,593 1 0 MATH-152 17.304 4,139 1 0 MATH-154 3.089 2,936 1 0.079 MATH-254A 11.353 1,000 1 0.001 MATH-4 9.816 723 1 0.002 MATH-5A 12.188 578 1 0 Table 1: Chi-Square analysis tutored vs non-tutored female students. Male Students Course Name Chi-Square N df Sig. (2-sided) MATH-12 1.254 1,659 1 0.263 MATH-152 18.303 3,943 1 0.000 MATH-154 13.847 2,730 1 0.000 MATH-254A 21.522 712 1 0.000 MATH-4 5.644 1,053 1 0.018 MATH-5A 2.790 1,067 1 0.095 Table 2: Chi-Square analysis tutored vs non-tutored male students. URM Students Course Name Chi-Square N df Sig. (2-sided) MATH-12 3.146 1,637 1 0.076 MATH-152 4.842 3,802 1 0.028 MATH-154 3.174 2,901 1 0.075 MATH-254A 26.060 1,103 1 0.000 MATH-4 3.481 634 1 0.062 MATH-5A 5.446 550 1 0.020 Table 3: Chi-Square analysis tutored vs non-tutored URM students. 14