Susan Dynarski, Brian Jacob, and Daniel Kreisman, "How Important Are Fixed Effects and Time Trends in Estimating Returns to Schooling? Evidence From A Replication of Jacobson, Lalonde and Sullivan, 2005", Journal of Applied Econometrics, Vol. 33, No. 7, 2018, pp. 1098-1108. The data for this project are subject to restricted access resulting from a research relationship between the University of Michigan and several unnamed community colleges. For this reason, the data cannot be made publicly available. For inquires about data access, please contact the University of Michigan Education Policy Initiative. For specific questions concerning the code or content, please contact Daniel Kreisman (dkreisman [AT] gsu.edu). We estimate 6 regressions in the main table (Table 3). We re-estimate the last 3 of these (columns 4-6 of Table 3) again by gender in Table 4. Then in the final table (Table 5), we re-estimate the last column of Table 3 for reference, and add two additional specifications. The .do files we used are in the file JLS_replication.do. Below we describe the sample limitation and give definitions of all the variables used in the regression models in the paper. These same variables are used in the summary statistics table, so we do not describe them again. 0.1 Sample Limitation Variable id match is the unique person identifier in the data. -- Drop observations after quarter 2, 2011. Variable quarter is defined by year-quarter (for example, 20112 is q2, 2011). Variable quarter tq is same in Stata %tq time. -- Keep if indicator ccfirst=1 -if first observed college enrollment is in our of our 5 community colleges. -- Keep those enrolled for the first time between 2002 and 2007. Variable first term enrolled is defined where terms are year-term; terms are defined over (3=spring, 5=summer, 7=fall, 9=winter; e.g. 20057 is fall, 2005). -- Drop observations where age or gender is missing. -- Keep those first enrolled between ages 21 and 45 (age first att credit is age first attempting a credit). -- Limit sample to observations when individuals are between ages 17 and 65. -- Limit to those not going on to a 4-year school (ever 4year = 1 if ever enrolled in a 4-year school, and (hdc ever is highest degree ever which = 4 if this is a BA). 0.2 Regression Tables We crease several sets of globals to enter the tables that correspond to the indicators at the bottom of Table 3. We define each in turn here. 1. Earnings measures and enrollment: -- Dependent variable is ui quarterly earnings real, which are real quarterly earnings from Unemployment Insurance records from the Michigan Department of Licensing and Regulatory Affairs. -- On their recommendation, values below $10 were recoded to $0. 2. Main Award variables (degree completion and timing). These define the highest degree student i ever earned as of quarter t. The term "unified" means highest degree at any school, not just our five institutions. Students who will go on to have a highest degree as a BA are dropped. Each of these are dummies with the omitted group never earning a degree. -- hdc_short_cert_unified: Highest degree as of quarter t is a short certificate (fewer than 15 credits). -- hdc_cert_unified: Highest degree as of quarter t is a certificate (more than 15 credits). -- hdc_assoc_unified: Highest degree as of quarter t is an Associates degree. 3. X_i vector, which are time-invariant characteristics: -- i.school_age_cohort: Full set of interactions between which of our 5 colleges i enrolled in, dummies for cohort of first enrollment in our colleges (semester), and dummies for age (in years) at first enrollment. -- i.race_male: Full set of interactions between gender and race (white/non-white). -- i.pell_loan: Full set of interactions between dummies for ever receiving a Pell grant, or ever taking a loan while enrolled in our schools. -- non_remedial_cratt_first, remedial_cratt_first: Linear term for number of remedial and non-remedial credits taken (attempted) in first term enrolled. -- mathZ, englishZ, has_score: First two are Z-scores for test used for english or math remediation. has_score is a dummy for whether i has test scores. 4. W_{it} vector, which are time-varying characteristics: -- age age2: Age and squared term in each quarter. • enrolled cc, enrolled nsc: Dummies for enrolled in one of our 5 colleges (enrolled cc) or enrolled in an NSC school, not including our 5 (enrolled nsc). -- hdc_*_enr_cc, hdc_*_enr_nsc : Interactions between time-invariant indicators for highest degree i ever completed and time-varying indicators for whether i is enrolled in one of our 5 colleges ( cc) or an NSC college ( nsc). -- enr_m1 enr_m2 enr_m3 enr_m4: These are dummies equal to 1 for the 4 quarters prior to enrollment. 5. X_{it} vector, which are interactions between time invariant characteristics and a linear time trend. -- c.trend: Linear term in quarters. -- This is interacted with all elements of the X_i vector. 6. post, post_trend, post_trend x degree. -- post: A dummy equal to 1 in all quarters after receiving highest degree (for degree earners) or in all quarters after last enrollment for non-completers. -- post_trend: Defined as the reciprocal of quarters since exit. -- pt_hdc_short_cert_unified_v2, pt_hdc_cert_unified_v2, and pt_hdc_short_cert_unified_v2: are interactions between the post-trend and which degree was earned. 7. \alpha_i is a person specific dummy. In the computation we use Stata's absorb command. 8. \tau_i are i.quarter_tq: Secular quarter dummies. 9. \omega_i \tau is a person fixed effect interacted with a linear time trend. We estimate this using Stata’s regintfe command as cited in the paper. 10. Matching quarters to semesters. To match quarters to semesters in order to determine enrollment we do the following. -- If student i was enrolled only in the spring semester (January through May), we defined enrollment in quarter 1 (January-March) equal to 1 (meaning enrolled the entire quarter), and enrollment in quarter 2 (April-June) equal to 2/3 (meaning i was enrolled 2/3rds of the quarter). -- If i was enrolled only in the summer semester (June-August), we defined enrollment in quarter 2 as 1/3 and enrollment in quarter 3 (July-Sept.) as 2/3. -- If i was enrolled only in the fall semester (Sept.-Dec.), we defined enrollment in quarter 3 as 1/3, and enrollment in quarter 4 (Oct.-Dec.) as 1. -- If i was enrolled in consecutive semesters, we would create combinations. For example, enrolling in fall and summer (January-August), enrollment in quarters 1 and 2 would equal 1, and enrollment in quarter 3 would equal 2/3. 11. Higher order trends, and pre-and post-trends. In table 5 we interact our person fixed effect with higher order trend (trend2) and with specific pre- and post-trends. -- The pre-trend is quarters relative to earning a degree or relative to last quarter enrolled. For example, a pre-trend value of 10 would correspond to the quarter 10 quarters prior to the last quarter i was enrolled. -- The post-trend is number of quarters since last enrollment or since earning the degree. For example, a value of 5 would correspond to the 5th quarter after earning an award. 0.3 Creating the Plots The plots are created by generating a measure that is quarters relative to both enrollment and college exit. This variable is negative in quarters until first enrollment (e.g. a value of -10 would correspond to 10 quarters prior to first enrollment) and is positive in quarters since last enrollment or since earning a degree (e.g. a value of 5 would correspond to 5 quarters post degree). This measure takes no value (i.e. is missing) in all quarters after first enrollment and before college exit. -- We then collapse earnings (mean and median) by highest degree ever earned (hdc ever) and our variable which is quarters relative to enrollment/exit. -- We then plot means by quarters relative to schooling across degree earner categories. 0.4 Summary Statistics Table This is simply created using the variables described in the regression section. Two additional measures are included which we describe here. -- prior emp is whether i was employed in the 1 year prior to enrolling. -- prior earnings 1yr is i's earnings in the year prior to enrollment.