This provides a brief tour of the replication files in the order in which they should be used:

1. replication_create_placebo.do

This .do file has the program that identifies the 44 metro areas that are defined in the 1977-1980 March CPS files. It creates a variable "demp" giving the city's rate of change of employment, and a variable "demp1" giving the city's rate of change in low-skill employment. It ranks the 44 cities by demp; this ranking appears in Table 3. The program creates a STATA dataset called goodsmsa.dta that contains the IPUMS "metarea" code for the 44 metros, and a dummy variable "good" set equal to 1, indicating this is a metro area that can be tracked throughout.


2. replication_marchcps.do

This .do file has the program that takes the downloaded version of the March CPS from IPUMS, selects the subsample of the 44 metro areas, and constructs the sample extract of working men 25-54. It also runs the year-specific regressions that give the age-adjusted log weekly wage. The program creates the raw data for the graphs presented in Figure 2 as well as the regressions reported in Table 4. Most of the other results in the paper pertaining to the March CPS can be obtained by additional manipulations of this file.


3. smsacode.xlsx and smsacode.dta

The ORG files change the code that identifies a metro area every few years. The excel file gives the information required to go from the ORG MSA codes to the IPUMS "metarea" variable. The smsacode.dta program is a STATA version of this data.


4. replication_cpsorg.do

This .do files takes the ORG STATA data sets downloaded from the NBER. It selects the correct number of metros by merging the ORG MSA codes with the data in smsacode.dta. It ensures that the education variable is defined consistently (the ORG codes change every few years). It then selects the sample of working men aged 25-64. It runs the year-specific regressions that create the age-adjusted log weekly wage. Finally, it constructs the figures that appear in Figure 4, and runs the regressions reported in Table 6. Most other ORG results in the paper can be obtained by further manipulations of this file.


5. experiment_org.do

This do file starts from scratch to see if I can replicate the ORG results. The ORG changes codes frequently, but the codes are stable between 1979 and 1985. This program just takes those 7 annual files, pools them, applies simple sample selection rules and sees if it can replicate the data for the same period using the more complex merging program that is required when we analyze the ORG over a longer period.



Note that the programs often "draw" graphs for the entire 1977-2001 period. The paper restricts the figures to 1977-1992.

