Evaluation of Teach For America:

EA15-536-2 Evaluation of Teach For America: 2014-2015 Department of Evaluation and Assessment Mike Miles Superintendent of Schools

This page is intentionally left blank. ii

Evaluation of Teach For America: 2014-2015 Approved Report of the Department of Evaluation and Assessment EA15-536-2 January 2015 This report was originally prepared for Commit! and the Texas Instruments Foundation. The report has been updated to add abstract and summary/recommendation sections. Nolan Rett Mickelson Amber McEnturff Joan Bush Manager Program Evaluation Nancy Kihneman, Ph.D. Director Program Evaluation Cecilia Oakeley, Ph.D. Assistant Superintendent Evaluation and Assessment Mike Miles Superintendent of Schools iii

This page is intentionally left blank. iv

Table of Contents ABSTRACT... 1 PURPOSE AND SCOPE OF THE EVALUATION... 1 MAJOR EVALUATION QUESTIONS AND RESULTS... 2 How many first-year teachers were in the Dallas ISD each year, and how many of these were TFA teachers?... 2 Methodology... 2 How does classroom effectiveness of TFA teachers compare with other first-year teachers in the district?... 3 Methodology... 3 Results... 4 How does the STAAR performance of students of TFA teachers compare with students of other teachers in the district?... 7 Methodology... 7 Results... 8 How does the retention of TFA teachers compare with non-tfa teachers in the district over time?... 25 Methodology... 25 Results... 25 SUMMARY AND RECOMMENDATIONS... 28 Summary... 28 Recommendation... 29 v

List of Tables Table Page 1 Number of First-Year Teachers and TFA teachers by School Year Cohort... 2 2 Number of Teachers with Valid CEI Scores by Cohort and Subject... 4 3 Logistic Regression Final Model Results for STAAR Mathematics Grades 4 to 5... 12 4 Logistic Regression Final Model Results for STAAR Mathematics Grades 6 to 8... 13 5 Logistic Regression Final Model Results for STAAR Algebra I End-of-Course Exam... 13 6 Logistic Regression Final Model Results for STAAR Reading Grades 4 to 5... 17 7 Logistic Regression Final Model Results for STAAR Reading Grades 6 to 8... 17 8 Logistic Regression Final Model Results for STAAR English I End-of-Course Exam... 18 9 Logistic Regression Final Model Results for STAAR English II End-of-Course Exam... 19 10 Logistic Regression Final Model Results for STAAR Science Grade 5... 22 11 Logistic Regression Final Model Results for STAAR Science Grade 8... 22 12 Logistic Regression Final Model Results for STAAR Biology End-of-Course Exam... 23 13 Logistic Regression Final Model Results for STAAR Social Studies Grade 8... 25 vi

List of Figures Figure Page 1 Average CEIs by Cohort and Content Area for TFA vs. Non-TFA Teachers... 5 2 Percentage That Met Expectations on Spring 2014 STAAR for Students of First-Year TFA and Non-TFA Teachers... 9 3 Percentage That Met Expectations on Spring 2014 STAAR for Students of Second-Year TFA and Non-TFA Teachers. 9 4 Percentage That Met Expectations on Spring 2014 STAAR for Students of Third- through Fifth-Year TFA and Non-TFA Teachers 9 5 Percentage That Met Expectations on STAAR Mathematics for Students of TFA and Non-TFA Teachers (Grades 3 to 5) 10 6 Percentage That Met Expectations on STAAR Mathematics for Students of TFA and Non-TFA Teachers (Grades 6 to 8)... 11 7 Percentage That Met Expectations on STAAR Algebra 1 EOC for Students of TFA and Non-TFA Teachers. 11 8 Percentage That Met Expectations on STAAR Reading for Students of TFA and Non-TFA Teachers (Grades 3 to 5)... 14 9 Percentage That Met Expectations on STAAR Reading for Students of TFA and Non-TFA Teachers (Grades 6 to 8)... 15 10 Percentage That Met Expectations on STAAR English 1 EOC for Students of TFA and Non-TFA Teachers.... 15 11 Percentage That Met Expectations on STAAR English 2 EOC for Students of TFA and Non-TFA Teachers... 16 12 Percentage That Met Expectations on STAAR Science for Students of TFA and Non-TFA Teachers (Grade 5)..... 20 13 Percentage That Met Expectations on STAAR Science for Students of TFA and Non-TFA Teachers (Grade 8).. 20 14 Percentage That Met Expectations on STAAR Biology EOC for Students of TFA and Non-TFA Teachers. 21 15 Percentage That Met Expectations on STAAR Social Studies for Students of TFA and Non-TFA Teachers (Grade 8)... 24 16 2009-10 Cohort Retention Rate by Year Began... 26 17 2010-11 Cohort Retention Rate by Year Began... 27 18 2011-12 Cohort Retention Rate by Year Began... 27 vii

19 2013-14 Cohort Retention Rate by Year Began... 28 20 2014-15 Cohort Retention Rate by Year Began... 28 viii

EVALUATION OF TEACH FOR AMERICA: 2014-2015 Project Evaluators: Nolan Rett Mickelson Amber McEnturff ABSTRACT Teach For America (TFA) is a program that recruits recent college graduates and prepares them to teach for two years in public schools. Since the 2009-2010 school year, 563 TFA teachers have worked in the Dallas Independent School District (ISD). Analyses of variance comparing the effectiveness of each cohort of TFA teachers in their first and second years to other teachers in their first or second years with the Dallas ISD indicated that most TFA cohorts were more effective, on average, than the comparison groups for both years. Several differences were statistically significant in mathematics and science. Using data from the 2013-14 school year in a series of logistic regressions, the evaluators assessed the impact of various teacher and student characteristics on students mathematics, reading, science, and social studies STAAR passing rates. For mathematics, having a TFA teacher was associated with significantly higher odds of passing across all grade levels after adjusting for student characteristics and teacher experience; results varied for other subjects and grade levels. TFA interns have returned to teach in the Dallas ISD for a second year at very high rates. As expected, a greater proportion of non-tfa than TFA teachers remained in district classrooms for three years or longer. PURPOSE AND SCOPE OF THE EVALUATION Teach For America (TFA) is a program that recruits high-achieving recent college graduates and prepares them to teach for two years in urban and rural public schools. TFA teachers have worked in the Dallas Independent School District (ISD) since the 2009-2010 school year. The purpose of this report is to summarize the classroom effectiveness and retention of TFA teachers compared with other teachers in the district and to review the performance of their students on the State of Texas Assessments of Academic Readiness (STAAR). This report was originally produced for Commit! and the Texas Instruments Foundation. It has been updated to include abstract and summary/recommendation sections. 1

MAJOR EVALUATION QUESTIONS AND RESULTS How many first-year teachers were in the Dallas ISD each year, and how many Methodology of these were TFA teachers? The Dallas ISD s Human Capital Management Department provided a list of TFA teachers with their hire dates. The group of first-year non-tfa teachers included: alternative certification (AC) interns who were part of the Dallas ISD s AC program, AC interns who were part of AC programs outside the district, certified teachers who had recently graduated from a traditional college or university program, and certified teachers with prior experience teaching in other Texas districts. Cohort was defined as the first school year an employee was a teacher with the Dallas ISD. 1 Limitations Sample size in subgroups was a limitation throughout the analyses. In some cases, comparisons between TFA and non-tfa teachers were not possible due to insufficient sample sizes. For analyses of student outcomes, only subgroups with at least two teachers were included. Though student sample sizes were typically large enough to make statistical conclusions, a large number of students sometimes represented a very small number of teachers. Therefore, results of subgroup analyses with small teacher sample sizes should be interpreted with caution. Upon consultation with representatives from TFA, 22 TFA teachers were discovered that were not included in the Dallas ISD s list of first-year teachers. These teachers were evenly distributed across content areas and grade levels, therefore not representing a large portion of any particular subgroup. The corrections from TFA were not received in time to update the current analysis but were documented to include in future TFA analyses. Results Table 1 shows the number of first-year teachers in the Dallas ISD by school year cohort. The number of first-year teachers and first-year TFA teachers increased over time. Table 1: Number of First-Year Teachers and TFA Teachers by School Year Cohort School Year Total Number of 1 st Year Teachers Number of 1 st Year TFA Teachers Percentage (%) of Total 2009-10 566 68 12.0 2010-11 707 97 13.7 2011-12 496 43 8.7 2012-13 1,363 84 6.2 2013-14 1,898 168 8.9 2014-15 1,124 103 9.2 Note: Number of TFA teachers included only TFA teachers found in the list of first-year teachers. 1 Cohort was determined using the teacher s hire date into the Dallas ISD. For example, a teacher hired during the summer preceding the 2013-14 school year or anytime during the 2013-14 school year would be included in the 2013-14 cohort. 2

How does classroom effectiveness of TFA teachers compare with other first-year teachers in the district? Methodology The evaluator conducted a series of Analyses of Variance (ANOVAs) to determine if differences in teaching effectiveness existed between TFA interns and all other teachers new to the Dallas ISD each year from 2009-10 to 2013-14. Analyses were conducted separately for Classroom Effectiveness Indices (CEIs) in Language Arts, Mathematics, and Science for both the first and second years of teaching. 2 CEIs indicate the relative effectiveness of a teacher after adjusting for student ethnicity, language proficiency, gender, socio-economic status, and academic performance from the previous year. Each year, CEIs are calculated to have a mean of 50 and standard deviation of 10. Because the first four cohorts year 1 and year 2 effectiveness comparisons were discrete and included the same teachers in both comparisons increasing the likelihood of finding a significant result when one may not exist the evaluator selected a statistical significance level of.025 for each statistical test (.05/2). The last cohort, 2013-14, had one year of CEIs available and maintained a significance level of.05. The evaluator measured effect size by calculating Eta-squared (η 2 ), a measure of the practical influence that the type of teacher (TFA or non-tfa) had on relative classroom effectiveness. Table 2 shows the number of TFA and Non-TFA first-year teachers with a valid CEI score for the first and second years of teaching by subject and cohort. 3 2 CEI scores are available for Foreign Language and Computer Science, but the extremely low number of teachers with scores on those indices would not allow for valid comparisons. 3 CEI scores measure teachers effectiveness relative to other teachers in the district each year and cannot be compared from year to year. Year 1 and year 2 CEIs were available for each cohort except the 2013-14 cohort, which had recently begun a second year in the Dallas ISD. 3

Table 2: Number of Teachers with Valid CEI Scores by Cohort and Subject Cohort and Year 1 Year 2 Subject 2009-10 Language Arts 23 93 25 107 Mathematics 18 110 21 132 Science 5 32 5 34 2010-11 Language Arts 33 110 30 127 Mathematics 31 109 28 140 Science 10 53 14 50 2011-12 Language Arts 12 68 10 78 Mathematics 14 67 13 76 Science 11 21 10 28 2012-13 Language Arts 24 179 28 218 Mathematics 25 173 23 225 Science 22 85 26 141 2013-14 Language Arts 60 447 -- -- Mathematics 59 379 -- -- Science 44 236 -- -- Note: The number of teachers receiving CEI scores may vary from year 1 to year 2 within each cohort due to teacher attrition and the number of teachers meeting the criteria for CEI calculation. Results The results from all 27 CEI comparisons are found in Figure 1. In general, TFA teachers were more effective than non-tfa teachers in teaching language arts, mathematics, and science in both their first and second years. No differences among teacher-types were statistically significant in language arts. For all three subjects, the average effectiveness of both TFA and non-tfa teachers in the first four cohorts (2009-10 through 2012-13) improved relative to all other teachers in the district during the second year of teaching. Language Arts In Language Arts, first-year TFA teachers generally received slightly higher first-year CEI scores than their non-tfa counterparts across all cohorts except the most recent, 2013-14 cohort (see Figure 1). None of the comparisons were statistically significant. No first-year TFA or non-tfa average scores eclipsed 50, the average for all district teachers. Second-year TFA teachers had slightly higher CEI scores than non-tfa second-year teachers with one exception: Non-TFA teachers in the 2009-10 cohort scored higher, on average, in their second year than TFA teachers (M = 49.9 and M = 48.9, respectively). The TFA second-year teachers from the 2011-12 (M = 50.0) and 2012-13 (M = 50.6) cohorts scored at or above the district mean of 50. As with the first year, differences among second-year teachers average Language Arts CEIs were not statistically significant. 4

Figure 1: Average CEIs by Cohort and Content Area for TFA vs. Non-TFA Teachers Mean CEI Score (2009-10) 60 36 Language Arts Mathematics Science 47.7 48.9 45.8 First 49.9 Second 51.5 46.3 First 55.7 50.0 Second* 55.7 47.6 46.7 48.2 First Second Mean CEI Score (2010-11) 60 36 47.9 48.9 47.1 48.4 First Second 53.2 52.7 49.9 46.2 First* Second 55.5 52.6 47.0 47.8 First Second* 60 57.7 Mean CEI Score (2011-12) 36 46.2 45.0 First 50.0 47.1 Second 49.7 46.9 First 52.2 48.4 Second 50.8 39.6 First* 46.6 Second* Mean CEI Score (2012-13) 60 36 49.4 46.5 First 50.6 48.8 Second 50.2 46.6 First 55.5 49.4 Second* 54.8 44.9 First* 56.6 51.1 Second* 60 Mean CEI Score (2013-14) 36 45.0 First 46.5 49.8 First* 46.1 50.8 First* 46.4 TFA Non-TFA Year Teaching (First or Second) Note: *Difference between TFA and non-tfa teachers was statistically significant (p <.025). 5

Mathematics As indicated in Figure 1, TFA teachers mean CEI scores in Mathematics were consistently higher than other first-year teachers across all five cohorts, ranging from 49.7 in 2011-12 to 53.2 in 2010-11. During the 2010-11 school year, the average Mathematics CEI score for TFA interns was 7.0 points higher than the average CEI score for other teachers. This result was statistically significant, F(1, 138) = 10.58, p =.001, and of medium effect (η 2 =.07). TFA interns who began teaching during the 2013-14 school year averaged 3.7 points higher than other first-year teachers, also a statistically significant result, F(1, 136) = 7.75, p =.006. The effect of teacher-type on effectiveness was small in this instance (η 2 =.02). The average Mathematics CEI scores for TFA interns in their second year were higher than other second-year teachers across all cohorts. The results were statistically significant for the 2009-10 cohort, F(1, 151) = 5.11, p =.025. The effect size, calculated as η 2, was equal to.03. This indicated that the type of second-year teacher (TFA or non-tfa) had a small to moderate effect on effectiveness. Results for the 2012-13 cohort s second year were also statistically significant, F(1, 246) = 8.35, p =.004, indicating a small to moderate effect (η 2 =.03). Every TFA cohort s average CEI score was above the overall district s average teacher score of 50, ranging from 52.2 in 2012-13 to 55.7 in 2009-10. Science TFA interns were generally more effective teaching science in their first year than other types of teachers new to the Dallas ISD, especially in the three most recent cohorts. All cohorts except the first, 2009-10 TFA cohort, averaged CEI scores above 50, meaning they were, as a group, more effective than most teachers in the Dallas ISD. With the exception of the 2009-10 cohort, TFA interns received higher Science CEIs on average than non-tfa teachers in their first year (see Figure 1). 4 First-year ANOVA results were statistically significant for the 2011-12, 2012-13, and 2013-14 cohorts. During the 2011-12 cohort s first year, TFA interns averaged 11.2 points higher than non-tfa teachers, F(1, 30) = 9.10, p =.005. This difference indicated a large teacher-type effect (η 2 =.23) on classroom effectiveness. TFA interns who began teaching with the 2012-13 cohort scored 9.9 points higher than other first-year teachers on average, F(1, 105) = 20.09, p <.001, with the effect size (η 2 ) again indicative of a large teacher-type effect by accounting for over 16 percent of the variance in effectiveness. The most recent 2013-14 cohort averaged 4.4 points higher in science than other teachers beginning that year, F(1, 278) = 7.54, p =.006. The influence of teacher-type on effectiveness, for this cohort, could be considered small to moderate (η 2 =.03). The second year of teaching for these cohorts yielded results similar to the first year. Across the four second-year TFA cohorts, average Science CEI scores were above the district overall average of 50, ranging from 55.5 to 57.7. The three most recent TFA cohorts to complete a second year 2010-11, 2011-12, and 2012-13 averaged effectiveness scores in science that were statistically significantly 4 It should be noted that the number of interns with valid Science CEIs in the 2009-10 cohort was small (n = 5); the difference was not statistically significant and should be reviewed with caution (see Table 2). 6

higher than their non-tfa second-year counterparts. The 2010-11 TFA interns averaged 7.7 points higher than other teachers, F(1, 62) = 7.79, p =.007, a result that is associated with a medium to large effect size (η 2 =.11). The 2011-12 TFA interns scored, on average, 11.1 points higher than other second-year teachers, F(1, 36) = 6.88, p =.013. The large effect size (η 2 =.16) indicated that type of second-year teacher accounted for over 16 percent of the variance in science classroom effectiveness. During the 2012-13 cohort s second year, TFA interns averaged 5.5 points higher than other teachers, F(1, 165) = 6.07, p =.015. Teacher-type, in this cohort s second year, had a moderate effect on effectiveness (η 2 =.04). How does the STAAR performance of students of TFA teachers compare with Methodology students of other teachers in the district? Students in grades three through eleven completed STAAR assessments in mathematics, reading, science, and social studies, and STAAR End-of-Course (EOC) assessments in Algebra I, English I and II, U.S. History, and Biology. 5 Students were linked to teachers using the Dallas ISD s course files, which listed the courses each student takes and the employee ID of the teacher. The employee ID of each student s teacher was used to label which 2013-14 students had teachers in their first five years of service and whether those teachers were TFA or not. 6 The course file was then merged with STAAR results to determine if there were differences in STAAR performance between students of TFA teachers and students of non-tfa teachers. 7 Two analyses were completed for each subject area and grade level. First, the percentage of students achieving a met expectations rating on STAAR was compared between TFA teachers and non-tfa teachers in their first, second, or third- through fifth-years of teaching. 8 A chi-square test of independence was used to assess whether the difference in the percentages was statistically significant. Second, hierarchical logistic regression was used to statistically test the impact of having a TFA teacher on achieving a met expectations rating on STAAR. The dependent variable was a dichotomous indicator of 1 if a student achieved a met expectations rating on 2013-14 STAAR and 0 if not. The predictor variables included prior achievement ( met expectations rating on 2012-13 STAAR), ethnicity, sex, 5 The general STAAR testing schedule is available here: http://tea.texas.gov/student.assessment/staar/ 6 Only teachers in the first five years of service were included because that is how long TFA teachers have been employed in the district. 7 To control for the possibility that TFA teachers were placed in lower-performing schools than non-tfa teachers, only campuses with TFA teachers were included in each analysis. For example, if elementary mathematics TFA teachers were found in five campuses, the comparison group was composed of non-tfa teachers at those same campuses. 8 TFA and non-tfa teachers were compared at different levels of experience: first-year teachers, second-year teachers, and third- through fifth-year teachers. Teachers in the third- through fifth-years of experience were grouped together due to small sample sizes. Subgroups were composed of different samples of teachers, not one group of teachers tracked over time. STAAR results came from the 2013-14 school year, so results were divided among those who were first-year teachers in 2013-14, those who were second-year teachers in 2013-14, etc. Therefore, use caution not to interpret differences between first-year and second-year teachers as improvement because the two groups are not composed of the same teachers. 7

socioeconomic status (SES), teacher level of experience, and whether the teacher participated in TFA. There were three models tested. In the first model, 2013-14 STAAR performance was predicted using only student characteristics. In the second model, teacher years of experience was included to see if including years of experience significantly improved the prediction of 2013-14 STAAR performance. In the third model, TFA was included. If the model including TFA did a significantly better job predicting 2013-14 STAAR performance, that indicated that having a TFA teacher made a difference over and above all the other factors considered (student characteristics, prior achievement, and teacher level of experience). Results STAAR Performance Overview Figures 2 through 4 show STAAR performance for students of TFA teachers at various levels of experience compared with students of non-tfa teachers. These percentages were further compared to passing rates across all teachers at TFA campuses to offer perspective on how students of these early-career TFA and non-tfa teachers performed relative to students of all teachers at the same campuses. Comparing across the three levels of experience and content areas, students of second-year TFA teachers had higher passing percentages than students of first-year TFA teachers. For mathematics, there was a trend of TFA teachers outperforming non-tfa teachers in the second year of experience. This was also true for elementary science. However, this trend was less apparent in reading/language arts. Small sample sizes resulted in missing data and precluded comparisons in many subgroups, impeding trend detection. More detailed analyses of STAAR performance data by grade level and content area follow. 8

Figure 2: Percentage That Met Expectations on Spring 2014 STAAR for Students of First-Year TFA and Non-TFA Teachers Percent That Met Expectations on STAAR 78 78 74 64 60 60 54 55 55 59 55 56 56 51 55 54 50 45 48 495152 42 41 939092 65 596163 56 53 ES MS Alg I ES MS Eng I Eng II ES MS Bio MS Mathematics Reading/Language Arts Science Social Studies All Figure 3: Percentage That Met Expectations on Spring 2014 STAAR for Students of Second-Year TFA and Non-TFA Teachers Percent That Met Expectations on STAAR 75 70 64 55 55 56 86 78 61 55 5756 656764 69 63 55 54 52 36 81 63 95 9592 ES MS Alg I ES MS Eng I Eng II ES MS Bio MS Mathematics Reading/Language Arts Science Social Studies 62 53 All Figure 4: Percentage That Met Expectations on Spring 2014 STAAR for Students of Third- through Fifth-Year TFA and Non-TFA Teachers Percent That Met Expectations on STAAR 76 797878 62 55 56 48 70 5956 64 64 64 44 75 55 54 5352 ES MS Alg I ES MS Eng I Eng II ES MS Bio MS Mathematics Reading/Language Arts Science Social Studies 71 63 92 53 All Note: Some subgroups were excluded from these figures due to small teacher sample sizes (n < 2). District percentages reflect students only in schools with TFA teachers in each subgroup. ES = Elementary School. MS = Middle School. 9

Mathematics Figure 5 shows the percentages of elementary students that met expectations on STAAR mathematics. Percentages were compared between first-year, second-year, and third- through fifth-year TFA and non-tfa teachers. For all levels of experience, a statistically significant higher percentage of elementary students of TFA teachers than non-tfa teachers passed STAAR mathematics. Figure 5: Percentage That Met Expectations on STAAR Mathematics for Students of TFA and Non-TFA Teachers (Grades 3 to 5) 76% Number of Teachers and Students 64% Represented in the Analysis 62% 54% 55% 42% 1st year teachers Teachers 12 22 Students 349 814 2nd year teachers Teachers 6 9 Students 261 254 1st year teachers** 2nd year teachers* 3rd-5th year teachers* 3rd-5th year teachers Teachers 4 10 Students 115 265 Note: Asterisks indicate level of significance for chi-square test of independence *p <.05 and **p <.001. Figure 6 shows the percentages of middle school students passing STAAR mathematics. For first-year teachers, there was no difference between TFA teachers and non-tfa teachers in percentage of students passing (55%). For second-year teachers, a higher percentage of students of TFA teachers (75%) passed STAAR mathematics compared with students of non-tfa teachers (70%), but this difference was not statistically significant. Passing percentages could not be compared for third- through fifth-year teachers because only one TFA teacher was represented in that subgroup. 10

Figure 6: Percentage That Met Expectations on STAAR Mathematics for Students of TFA and Non-TFA Teachers (Grades 6 to 8) 75% Number of Teachers and Students 70% Represented in the Analysis 55% 55% 48% 1st year teachers Teachers 10 14 Students 1,053 1,346 2nd year teachers 1st year teachers 2nd year teachers n/a 3rd-5th year teachers 3rd-5th year teachers Teachers 6 6 Students 525 519 Teachers 1 9 Students 5 693 Note: These differences were not statistically significant. Percentage was not reported for students of third- through fifth-year TFA teachers because there was only one teacher represented in this subgroup. At the high school level, Figure 7 shows passing percentages on the STAAR Algebra I EOC exam. For first-year and third- through fifth-year teacher subgroups, there was no statistically significant difference between TFA teachers and non-tfa teachers in percentage of students passing. For second-year teachers, a higher percentage of students of TFA teachers (86%) passed STAAR mathematics compared with students of non-tfa teachers (55%). The difference was statistically significant. Figure 7: Percentage That Met Expectations on STAAR Algebra 1 EOC for Students of TFA and Non-TFA Teachers 86% Number of Teachers and Students 78% 79% 78% 74% Represented in the Analysis 55% 1st year teachers Teachers 14 4 Students 1,038 248 2nd year teachers Teachers 6 3 Students 472 152 1st year teachers 2nd year teachers* 3rd-5th year teachers 3rd-5th year teachers Teachers 2 3 Students 191 235 Note: Asterisk indicates level of significance for chi-square test of independence *p <.001. Logistic regression results for elementary STAAR mathematics are shown in Table 3. The results indicated that prior achievement, teacher experience level, and TFA status were associated with 11

significantly greater likelihood of passing STAAR mathematics. 9 The odds ratio for TFA status indicated that elementary students of TFA teachers were 3.228 times more likely to pass STAAR mathematics than students of non-tfa teachers when accounting for teacher experience and student characteristics. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 10 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), compared with Hispanic students, students who were African American (odds ratio = 0.540) or Other Ethnicity (odds ratio = 0.159) were significantly less likely to pass STAAR. According to the odds ratio, students who passed STAAR mathematics in 2012-13 were 13.112 times more likely to pass STAAR in 2013-14. Compared with first-year teachers, students of second-year teachers were 2.179 times more likely to pass STAAR, and students of third- through fifth-year teachers were 2.645 times more likely to pass STAAR. Table 3: Logistic Regression Final Model Results for STAAR Mathematics Grades 4 to 5 Predictor B z Odds Ratio STAAR Mathematics Met Expectations 2012-13 2.574 282.910** 13.112 African American -0.615 12.898** 0.540 Other Ethnicity -1.838 4.048* 0.159 Male 0.084 0.344 - Low SES -0.712 2.671 - Second-year teacher 0.779 19.067** 2.179 Third- through fifth-year teacher 0.973 25.003** 2.645 TFA 1.172 48.693** 3.228 Note: Final model Nagelkerke R 2 =.431. *p <.05 and **p <.001. Grade 3 students in 2013-14 were omitted because they have no prior STAAR achievement rating. Logistic regression results for middle school STAAR mathematics are shown in Table 4. The results indicated that prior achievement, teacher experience level, and TFA participation were associated with significantly greater likelihood of passing STAAR mathematics. The odds ratio for TFA status indicated that middle school students of TFA teachers were 1.230 times more likely to pass STAAR mathematics than students of non-tfa teachers when teacher experience and student characteristics were held constant. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 11 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), compared with Hispanic students, students who were African American were significantly less likely to pass STAAR (about half as likely; odds ratio = 0.476) after holding teacher experience and student characteristics constant. According to the odds ratio, students who passed STAAR mathematics in 2012-13 were 8.205 times more likely to pass STAAR in 9 Odds ratios greater than one indicated the predictor was associated with greater likelihood of passing STAAR. Odds ratios less than one indicated the predictor was associated with less likelihood of passing STAAR. 10 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.392 (Model 2) to.431. Model fit improvement as measured by -2 log likelihood was statistically significant (p <.001). 11 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.331 (Model 2) to.332. Model fit improvement as measured by -2 log likelihood was statistically significant (p =.015). 12

2013-14 when teacher experience and student characteristics were held constant. Compared with firstyear teachers, students of second-year teachers were 2.169 times more likely pass STAAR. However, students of third- through fifth-year teachers were less likely than students of first-year teachers to pass STAAR (odds ratio = 0.765) when teacher experience and student characteristics were held constant. Table 4: Logistic Regression Final Model Results for STAAR Mathematics Grades 6 to 8 Predictor B z Odds Ratio STAAR Mathematics Met Expectations 2012-13 2.105 739.687** 8.205 African American -0.743 65.277** 0.476 Other Ethnicity -0.189 0.356 - Male -0.054 0.501 - Low SES -0.006 0.001 - Second-year teacher 0.774 65.241** 2.169 Third- through fifth-year teacher -0.269 6.036* 0.765 TFA 0.207 5.906* 1.230 Note: Final model Nagelkerke R 2 =.332. *p <.05 and **p <.001. Logistic regression results for STAAR Algebra I EOC are shown in Table 5. The results indicated that prior achievement and TFA status were associated with significantly greater likelihood of passing STAAR Algebra I EOC. The odds ratio for TFA status indicated that Algebra I students of TFA teachers were 1.630 times more likely to pass the Algebra I EOC exam than students of non-tfa teachers when teacher experience and student characteristics were held constant. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 12 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), students who passed STAAR mathematics in 2012-13 were 3.951 times more likely to pass STAAR in 2013-14 after holding teacher experience and student characteristics constant. Compared with Hispanic students, students who were African American were significantly less likely to pass the STAAR EOC (odds ratio = 0.719) when teacher experience and student characteristics were held constant. No other student or teacher characteristics were significant predictors of passing STAAR. Table 5: Logistic Regression Final Model Results for STAAR Algebra I End-of-Course Exam Predictor B z Odds Ratio STAAR Mathematics Met Expectations 2012-13 1.374 128.348** 3.951 African American -0.330 5.460* 0.719 Other Ethnicity 0.136 0.128 - Male -0.185 2.370 - Low SES 0.104 0.329 - Second-year teacher 0.180 1.532 - Third- through fifth-year teacher 0.303 2.919 - TFA 0.488 12.491** 1.630 Note: Final model Nagelkerke R 2 =.125. *p <.05 and **p <.001. 12 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.116 (Model 2) to.125. Model fit improvement as measured by -2 log likelihood was statistically significant (p <.001). 13

Mathematics Results Summary At the elementary school level, a greater percentage of students of TFA teachers of all experience levels passed the STAAR mathematics test. For Algebra I students, a greater percentage of students of second-year TFA teachers passed the Algebra I EOC exam. There were no other significant differences in passing rates between TFA and non-tfa teachers. According to the logistic regression results, having a TFA teacher significantly improved the odds of a student passing STAAR mathematics, ranging from odds 1.2 times higher for middle school students to more than three times higher for elementary students after holding teacher experience and student characteristics constant. Students of non-tfa teachers did not statistically significantly outperform students of TFA teachers in any of the comparisons reported above. Reading/Language Arts Figure 8 shows passing percentages on STAAR reading for students in grades three through five. For first-year and second-year teacher subgroups, there was no statistically significant difference between TFA teachers and non-tfa teachers in percentage of students passing. For third- through fifth-year teachers, a statistically significant higher percentage of students of TFA teachers (70%) passed STAAR reading compared with students of non-tfa teachers (59%). Figure 8: Percentage That Met Expectations on STAAR Reading for Students of TFA and Non-TFA Teachers (Grades 3 to 5) Number of Teachers and Students 70% Represented in the Analysis 61% 57% 59% 51% 50% 1st year teachers Teachers 19 33 Students 662 1,058 2nd year teachers Teachers 4 15 Students 173 481 1st year teachers 2nd year teachers 3rd-5th year teachers* 3rd-5th year teachers Teachers 4 14 Students 91 437 Note: Asterisks indicate level of significance for chi-square test of independence *p <.05. Figure 9 shows passing percentages on STAAR reading for students in grades six through eight. For all teacher subgroups, there was no statistically significant difference between TFA teachers and non-tfa teachers in percentage of students passing. 14

Figure 9: Percentage That Met Expectations on STAAR Reading for Students of TFA and Non-TFA Teachers (Grades 6 to 8) Number of Teachers and Students Represented in the Analysis 65% 67% 60% 60% 64% 64% 1st year teachers Teachers 9 33 Students 621 2,807 2nd year teachers Teachers 13 14 Students 1,020 767 1st year teachers 2nd year teachers 3rd-5th year teachers 3rd-5th year teachers Teachers 4 12 Students 301 689 Note: These differences were not statistically significant. Figure 10 shows passing percentages on the STAAR English I EOC assessment. There was no statistically significant difference between first-year TFA teachers and non-tfa teachers in percentage of students passing. However, the percentage of students passing was statistically significantly higher for students of third- through fifth-year non-tfa teachers (75%) than students of TFA teachers (44%). Figure 10: Percentage That Met Expectations on STAAR English 1 EOC for Students of TFA and Non-TFA Teachers Number of Teachers and Students Represented in the Analysis 75% 1st year teachers 45% 41% 44% 1st year teachers 2nd year teachers 3rd-5th year teachers* TFA n/a Non-TFA Teachers 4 5 Students 363 274 2nd year teachers Teachers 0 1 Students 0 133 3rd-5th year teachers Teachers 2 2 Students 200 201 Note: Asterisks indicate level of significance for chi-square test of independence *p <.001. Percentage is not reported for students of second-year teachers because there was only one non-tfa teacher and no TFA teachers represented in this subgroup. Figure 11 shows passing percentages on the STAAR English II EOC assessment. A significantly greater percentage of students of first-year TFA teachers passed the English II EOC (59%) than students of non-tfa teachers (48%). Sixty-three percent of students of non-tfa second-year teachers passed the English II EOC. Comparisons between TFA and non-tfa were not possible for second- or third- through fifth-year teachers due to small teacher sample sizes. 15

Figure 11: Percentage That Met Expectations on STAAR English 2 EOC for Students of TFA and Non-TFA Teachers Number of Teachers and Students Represented in the Analysis 59% 63% 1st year teachers 48% Teachers 4 5 Students 354 302 1st year teachers* n/a 2nd year teachers n/a 3rd-5th year teachers 2nd year teachers 3rd-5th year teachers Teachers 1 2 Students 93 209 Teachers 1 1 Students 52 118 Note: Asterisks indicate level of significance for chi-square test of independence *p <.05. Some percentages were not reported due to small sample sizes. Logistic regression results for elementary STAAR reading are shown in Table 6. The results indicated that prior achievement, teacher experience level, and TFA participation were associated with significantly greater likelihood of passing STAAR reading. 13 Elementary students of TFA teachers had 1.607 times higher odds of passing STAAR reading than students of non-tfa teachers after holding teacher experience and student characteristics constant. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 14 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), students who passed STAAR reading in 2012-13 were 11.869 times more likely to pass STAAR reading in 2013-14 when teacher experience and student characteristics were held constant. The odds of students of second-year teachers passing the STAAR were 1.585 times higher than the students of first-year teachers, but students of third- through fifth-year teachers were not significantly more or less likely to pass STAAR when holding teacher experience and student characteristics constant. Compared with Hispanic students, students who were African American were significantly less likely to pass STAAR (odds ratio = 0.623). 13 Odds ratios greater than one indicated the predictor is associated with greater likelihood of passing STAAR. Odds ratios less than one indicated the predictor is associated with less likelihood of passing STAAR. 14 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.376 (Model 2) to.383. Model fit improvement as measured by -2 log likelihood was statistically significant (p <.001). 16

Table 6: Logistic Regression Final Model Results for STAAR Reading Grades 4 to 5 Predictor B z Odds Ratio STAAR Reading Met Expectations 2012-13 2.474 413.799** 11.869 African American -0.474 13.533** 0.623 Other Ethnicity 0.765 2.087 - Male -0.222 3.498 - Low SES -0.270 0.621 - Second-year teacher 0.460 8.259* 1.585 Third- through fifth-year teacher -0.022 0.019 - TFA 0.474 12.222** 1.607 Note: Final model Nagelkerke R 2 =.383. *p <.05 and **p <.001. Grade 3 students in 2013-14 were omitted because they have no prior STAAR achievement rating. Logistic regression results for middle school STAAR reading are shown in Table 7. The results indicated that prior achievement and teacher experience level were associated with significantly greater likelihood of passing STAAR reading, while being classified as male and low SES were associated with less likelihood of passing. Having a TFA teacher was not a significant predictor of passing STAAR reading for middle school students. The addition of TFA status in Model 3 did not result in a statistically significant improvement in overall model fit, indicating that TFA status did not have an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa) and with teacher experience and student characteristics held constant, students who passed STAAR reading in 2012-13 were 9.298 times more likely to pass STAAR reading in 2013-14. Compared with first-year teachers, the odds of students of second-year teachers passing the STAAR were 1.324 times higher, and students of third- through fifth-year teachers were 1.249 more likely to pass STAAR when student characteristics were held constant. Middle school males were significantly less likely than females to pass STAAR reading (odds ratio = 0.655). Furthermore, students with low SES were less likely to pass STAAR reading (odds ratio = 0.621) when holding teacher experience and other student characteristics constant. Table 7: Logistic Regression Final Model Results for STAAR Reading Grades 6 to 8 Predictor B z Odds Ratio STAAR Reading Met Expectations 2012-13 2.230 1,232.991** 9.298 African American -0.119 1.628 - Other Ethnicity 0.294 2.425 - Male -0.423 44.136** 0.655 Low SES -0.476 15.498** 0.621 Second-year teacher 0.280 14.591** 1.324 Third- through fifth-year teacher 0.223 6.083* 1.249 Note: Final model Nagelkerke R 2 =.322. *p <.05 and **p <.001. TFA was not a significant predictor and was not included in the final model. Logistic regression results for the STAAR English I EOC exam are shown in Table 8. The results indicated that prior achievement and teacher experience level were associated with significantly greater likelihood of passing the STAAR English I EOC, while having a TFA teacher was associated with less likelihood of passing. The odds ratio indicated that students of TFA teachers were about half as likely to pass the English I EOC as students of non-tfa teachers (odds ratio = 0.491) when holding teacher 17

experience and student characteristics constant. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 15 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), students who passed STAAR reading in 2012-13 had 9.537 times higher odds of passing the STAAR English I EOC in 2013-14. Compared with first-year teachers, students of third- through fifth-year teachers were 1.463 times more likely pass STAAR when student characteristics were held constant. No other student or teacher characteristics were significantly associated with passing the English I EOC. Table 8: Logistic Regression Final Model Results for STAAR English I End-of-Course Exam Predictor B z Odds Ratio STAAR Reading Met Expectations 2012-13 2.255 138.101** 9.537 African American 0.085 0.220 - Other Ethnicity -0.250 0.371 - Male -0.229 2.671 - Low SES -0.101 0.211 - Second-year teacher -0.093 0.150 - Third- through fifth-year teacher 0.380 5.694* 1.463 TFA -0.711 21.421** 0.491 Note: Final model Nagelkerke R 2 =.273. *p <.05 and **p <.001. Logistic regression results for the STAAR English II EOC exam are shown in Table 9. The results indicated that prior achievement and teacher experience level were associated with significantly greater likelihood of passing the STAAR English II EOC, while being African American or male were associated with less likelihood of passing. Having a TFA teacher was not a significant predictor of passing the STAAR English II EOC exam. The addition of TFA status in Model 3 did not result in a statistically significant improvement in overall model fit, indicating that TFA status did not have an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), students who passed the STAAR Reading I EOC in 2012-13 were 10.056 times more likely to pass STAAR English II EOC in 2013-14 after holding other student characteristics and teacher experience constant. Compared with students of first-year teachers, students of third- through fifth-year teachers were 1.781 times more likely to pass STAAR. Students who were African American had lower odds of passing than students who were Hispanic (odds ratio = 0.417), and students who were male were about half as likely to pass as females (odds ratio = 0.563) when holding constant teacher experience and other student characteristics. No other student or teacher characteristics were significantly associated with passing the English II EOC. 15 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.250 (Model 2) to.273. Model fit improvement as measured by -2 log likelihood was statistically significant (p <.001). 18

Table 9: Logistic Regression Final Model Results for STAAR English II End-of-Course Exam Predictor B z Odds Ratio STAAR EOC Reading I Met Expectations 2012-13 2.308 234.116** 10.056 African American -.876 22.356** 0.417 Other Ethnicity.116 0.039 - Male -.575 14.942** 0.563 Low SES.285 1.588 - Second-year teacher.241 1.931 - Third- through fifth-year teacher.577 6.892* 1.781 Note: Final model Nagelkerke R 2 =.363. *p <.05 and **p <.001. Reading Results Summary At the elementary school level, students of third- through fifth-year TFA teachers had significantly higher passing rates than students of non-tfa teachers. In middle school, there were no significant differences between TFA and non-tfa teachers. Results were mixed on English I and II EOC exams. Students of first-year TFA teachers had better passing rates on English II, but students of third- through fifth-year non-tfa teachers had better passing rates on English I. Logistic regression results were also mixed. Having a TFA teacher improved the odds of passing STAAR reading for elementary and middle school students (odds ratios of 1.607 and 1.249, respectively) when controlling for teacher experience and certain student characteristics. However, there was no relationship between having a TFA teacher and passing the English II EOC, and students of TFA teachers were half as likely to pass the English I EOC as students of non-tfa teachers. Science Figure 12 shows passing percentages on STAAR science for students in grade five. There was no statistically significant difference between first-year TFA teachers and non-tfa teachers in percentage of students passing. However, the percentage of students passing was statistically significantly higher for students of second-year TFA teachers (69%) than students of non-tfa teachers (36%). Though 53 percent of students of third- through fifth-year non-tfa teachers passed STAAR science, there were no students of TFA teachers in this subgroup with which to compare. 19

Figure 12: Percentage That Met Expectations on STAAR Science for Students of TFA and Non-TFA Teachers (Grade 5) Number of Teachers and Students Represented in the Analysis 69% 1st year teachers 49% 51% 53% Teachers 5 6 36% Students 179 231 2nd year teachers 1st year teachers 2nd year teachers* n/a 3rd-5th year teachers 3rd-5th year teachers Teachers 2 4 Students 109 139 Teachers 0 5 Students 0 146 Note: Asterisks indicate level of significance for chi-square test of independence *p <.05. Some percentages were not reported due to small sample sizes. Figure 13 shows passing percentages on STAAR science for students in grade eight. There was no statistically significant difference between first-year TFA teachers and non-tfa teachers in percentage of students passing. For second-year TFA teachers, 81 percent of students passed STAAR science, while 71 percent of third- through fifth-year TFA teachers passed. There were no non-tfa second or third- through fifth-year teachers. Figure 13: Percentage That Met Expectations on STAAR Science for Students of TFA and Non-TFA Teachers (Grade 8) Number of Teachers and Students 81% Represented in the Analysis 71% 59% 61% 1st year teachers Teachers 4 2 Students 422 209 1st year teachers n/a 2nd year teachers n/a 3rd-5th year teachers 2nd year teachers 3rd-5th year teachers Teachers 2 0 Students 260 0 Teachers 2 0 Students 257 0 Note: These differences are not statistically significant. Some percentages were not reported due to small sample sizes. Figure 14 shows passing percentages on the STAAR Biology EOC. There was no statistically significant difference for first-year or second-year TFA teachers and non-tfa teachers in percentage of students passing. There were no non-tfa third- through fifth-year teachers to compare with TFA teachers. 20

Figure 14: Percentage That Met Expectations on STAAR Biology EOC for Students of TFA and Non-TFA Teachers 93% 90% 95% 95% Number of Teachers and Students Represented in the Analysis 1st year teachers Teachers 6 7 Students 660 586 2nd year teachers Teachers 7 4 n/a Students 595 408 1st year teachers 2nd year teachers 3rd-5th year teachers 3rd-5th year teachers Teachers 1 0 Students 79 0 Note: These differences are not statistically significant. Some percentages were not reported due to small sample sizes. Logistic regression results for STAAR science for grade five are shown in Table 10. The results indicated that prior achievement, teacher experience level, and TFA status were associated with significantly greater likelihood of passing STAAR science, while no predictors were associated with less likelihood of passing. The odds ratio for TFA status indicated that fifth-grade students of TFA teachers had 1.684 times higher odds of passing STAAR science than students of non-tfa teachers when holding teacher experience and student characteristics constant. The addition of TFA status in Model 3 resulted in a statistically significant improvement in overall model fit, 16 indicating that TFA status had an impact on STAAR performance over and above other student and teacher characteristics in the model. Across all teachers (TFA and non-tfa), students who passed STAAR mathematics in 2012-13 17 were 6.158 times more likely to pass STAAR science in 2013-14. Compared with first-year teachers, students of third- through fifth-year teachers were 1.632 times more likely pass STAAR when student characteristics were held constant. No other student or teacher characteristics were significantly associated with passing STAAR science at grade five. 16 Addition of TFA status in Model 3 increased Nagelkerke R 2 from.226 (Model 2) to.238. Model fit improvement as measured by -2 log likelihood was statistically significant (p <.01). 17 STAAR mathematics was used as an indicator of prior achievement for STAAR science in grades five and eight because no STAAR science assessment was administered to students in grades four and seven to serve as prior achievement. For Biology EOC, which is administered primarily to students in grade nine, grade eight STAAR science was used as prior achievement. 21