Exercise L3, linear growth model
Papke (1994) analysed data from
1980 to 1988 to establish the effectiveness of
Number of observations (rows): 198
Number of variables (columns): 41
Variables we use in this exercise:
city= city identifier (1,2,…,22)
year= calendar year (1980,1981,…,1988)
uclms=number of unemployment claims
t=linear time trend
ez=1 if the city is in the enterprise zone, 0 otherwise
d8m=1 if year is 198m, 0 otherwise, m=1,2,3,4,5,6,7,8
cm=1 if city=m.0 otherwise (m=1,2,…,22)
Start Sabre and specify transcript file:
data year uclms ez d81 d82 d83 d84 d85 d86 d87 d88 c1 c2 c3 c4 c5 c6 c7 c8 &
c9 c10 c11 c12 c13 c14 c15 c16 c17 c18 c19 c20 c21 c22 luclms t ezt &
(1) Estimate a linear model on the log of number of unemployment claims (luclms) without covariates.
(2) Allow for the city identifier (city) random effect use mass 64 and starting values sigma 0.5, scale 0.5. Is this random effect significant?
(3) Add the binary ez effect. How does the magnitude of the city random effect change? Is the enterprise zone effect significant in this model?
(4) Add the linear time effect (t), use starting value sigma 0.3. How does the magnitude of the city specific random effect change?
(5) Remove the linear time effect (t) and add the set of dummy variables (d8m, m=1,2,…,8) in its place, use starting value sigma 0.2. Do the d8m dummy variables work as well as the linear time effect? If not, why not?
(6) Interpret your preferred model, does ez have an effect on the response log(uclms)?
(7) Re-estimate your preferred model using the dummy variables for city (c1,c2,…,c22) instead of treating the city effect as random variables. How do the fixed effect model results differ from those of the random effects model?
Papke, L. E., (1994), Tax policy
and urban development: Evidence from the
Rabe-Hesketh, S., and Skrondal, A., (2005), Multilevel and Longitudinal Modelling using Stata, Stata Press, Stata Corp, College Station, Texas