We replicated Neumayer (2003)'s two-way fixed effects estimates using a panel of homicide data from up to 117 countries over the period 1980-97. Neumayer (2003) also ran dynamic panel data estimation using Arellano and Bond GMM; see column 3 of his Table III. He found insignificant lagged effects of homicide. However, zero first-order serial correlation was not rejected. This renders the use of lagged homicide rates as instruments invalid. The data set and Stata code are provided on the author's university web page. (http://www2.lse.ac.uk/geographyAndEnvironment/whosWho/profiles/neumayer/replicationdatasets2. aspx).

(a) Replicate column 3 of Table III of Neumayer (2003) which reports Arellano and Bond GMM. Perform the Sargan over-identification test and zero first and second-order tests for serial correlation. What do you conclude?

(b) Perform system GMM rather than Arellano and Bond GMM using xtdpdsys and the corresponding diagnostics. What do you conclude?