Jacques Melitz has found an error in my bilateral trade data sets, for which I thank him. In particular, the common language dummy variable (mnemonic “comlang”) is coded incorrectly for some Belgian and Swiss observations.
Switzerland was incorrectly coded as having English, French and German as its official languages instead of Italian, French and German. Observations for Belgium (also a tri-lingual country with Flemish/Dutch, French and German as official languages) were dropped, owing to the fact that TSUSA data is for the Belgium-Luxembourg economic union, which does not correspond to any PWT or WDI country. Thus Belgium does not share a common language with any country in the (flawed) data set.
To my knowledge, no other observations or variables are in error.
These data sets were used in my paper (with Charles Engel) “Currency Unions and International Integration” NBER WP # 7872, my paper (with Jeff Frankel) “An Estimate of the Effect of Currency Unions on Trade and Growth” NBER WP #7857, and my paper (with Feenstra and Markusen) “Understanding the Home Market Effect and the Gravity Equation: The Role of Differentiating Goods”, forthcoming Canadian Journal of Economics. I have regenerated the data sets, re-estimated the affected output and corrected the results (Tables 3 and 7 in my paper with Engel; Tables 1 and 2 in my paper with Frankel; Tables 2-4 in my paper with Feenstra and Markusen). The effects are small. Corrected results are available in the PDF files on my website, and also in forthcoming CEPR DPs (for my papers with Engel and Frankel). The data sets available for my papers with Engel, Frankel and Feenstra and Markusen are available and have been corrected.
The flaw also affects my “One Money, One Market: Estimating The Effect of Common Currencies on Trade.” Given the nature of the problem, I have decided not to publish an erratum. I have also left the original data set on my website, but urge users to either use the corrected version or to correct the data themselves.
I apologize for this mistake, and again ask future users of my data sets to bring any problems to my attention.
***
In April 2001, Patrick Honohan alerted me to two other mistakes in the data. First, Guinea-Bissau was not a member of the CFA franc zone until 1997 (I incorrectly record it as having been in from 1970 through 1990). Second, Barbados was a member of the ECCA in 1970 (I incorrectly record it as having not been). I thank Patrick for bringing these errors to my attention, and again apologize for my mistakes.
***
In May 2001, Romain Wacziarg found other mistakes associated with the coding of “comctry” for Suriname. In particular, Suriname is incorrectly coded as being part of the Netherlands for the entire sample (it was granted independence in 1975). Also, it is incorrectly coded as being part of the same country as Portugal (and hence Angola). I thank Romain for bringing these errors to my attention, and again apologize for my mistakes.
***
In May 2001, Volker Nitsch alerted me to the fact that New Zealand is coded as being a member of the EEC/EC/EU instead of the Netherlands. I thank him for bringing attention to this error, and again apologize.
***
In December 2001, Peter Kenen alerted me to the fact that Equatorial Guinea (country code 162260) is mis-labelled “Fr.So.Ant.Tr.” Also, the Korean exchange rate volatilities are actually those of Cote d’Ivoire. Sorry!
***
In March 2002, Charles Engel alerted me to the fact that the 1995 logs of the products of population and real GDP are a factor of 1,000,000 greater than those of 1970-1990 for my bilateral data set with Frankel. This turns out not to affect the results in the paper, and has been since been corrected in the posted version.
***
In June 2005, Dennis Novy noticed that some of the common colonizer dummies are incorrect (for India, Nigeria, and Hong Kong). Sorry! I urge users both to correct these mistakes or (preferably) to use the Glick-Rose data set which is, to my knowledge, error-free.
*****
On October 21, 2005 Russell Hillberry alerted me to the fact that EFTA is not included in the RTAs (as noted in the paper, so this is a clarification rather than an error). Thanks to him.
*****
On March 7, 2007 James Foreman-Peck alerted me to the fact that real trade volumes seem to decline “too much” during periods of inflation which are also periods of real trade expansion. The likely reason is a units issue with the deflator (100 instead of 1?), though I have not checked. This is not a problem for regressions so long as year effects are always included, but I urge users to be aware. Thanks to James for pointing this out.