Andrew Sweeting 
Associate Professor 
University of Maryland 
College Park, MD 20742
Matthew DelNero
Chief, Wireline Competition Bureau 
Federal Communications Commission 
Washington D.C. 20554
July 13, 2016
Dear Matthew
Comments on New Materials Related to the “Empirics of Business Data Services”
On July 1 you invited me to review and comment on some new materials made available after the 
reviews of Professor Rysman’s “Empirics  of Business Data Services” report.  These materials 
reflect changes made by cable companies to the data supplied, and some additional analysis that 
followed issued raised in two referee reports, including my own. Further revisions were provided 
on July 12. I have reviewed these materials, although in less detail than my review of the original 
Rysman study. I have a few comments that I am happy to place in the record.
(1) Clustering of standard errors: in my initial report I argued that it was important to check that 
the statistical significance of the regression results was robust to the “clustering” of standard 
errors, because it  was plausible that some common unobserved factors affect the prices 
charged by incumbent carriers in nearby buildings, which would mean that the observations 
were not truly independent of each other. I tentatively suggested the clustering at the census 
block level might be appropriate.
The results in the staff memorandum show that clustering at the block level can increase 
standard errors substantially (for example by a factor of 2 or 3) although this does not change 
whether the coefficients of interest are statistically significance, at the 5% level, in most 
cases.  While this is comforting in one sense, it would not be unreasonable for someone to 
argue that if allowing for clustering at the block level has such a large effect on standard 
errors, then it is possible that widening the geography even further, for example, to the tract 
level, might further increase the standard errors and start to change the significance of more 
coefficients.
(2) Cable facilities: I like the fact that the FCC staff have included new measures of cable 
presence, measured in a variety of ways, in a new analysis. One might have argued that these 
connections could provide some substitute competition to the type of “facilities based 
building measure” that was included in the previous regressions, especially as it seems that 
this often reflects actual connections rather than the ability to easily provide a connection. 
Depending on the correlation of the cable presence measures and the previous measure, one 
could have imagined that this might affect the estimates in different directions.
The new results themselves are a little bit hard to interpret, partly because there are so many 
regressions, and partly because these results might also be affected if the clustering was 
broadened.  We seem to see a slightly different pattern for different types of service.  For 
high-bandwidth connections there seems to be some evidence that the cable presence
measures reduce prices, and more restrictive measures of that competition seem to reduce 
prices slightly more (e.g., Table 3.1a). For DS-3 we see some large price-reducing effects of 
DOCSIS 3.0 availability.  But the effects seem to come primarily from “price cap 1” areas 
where some of the estimates are very large indeed.  In my original report I suggested that 
some measure of plausibility of price effects might come from engineering estimates of costs. 
Apparently this is not possible, so I do not have any clear basis to disagree with the staff 
assessment that such large  effects are likely to be picking up something else.  A simple 
analysis of summary statistics (whether in terms of demographics, business density, types of 
customers, and factors that might affect costs of service) might be useful to understand what 
is going on here.
I hope that these comments are useful. 
Yours sincerely
Andrew Sweeting