Ideological Segregation Online and Offline
The Quarterly Journal of Economics, 2011 · paper · Media studies
Measures ideological segregation of online news, offline news and face-to-face interactions in the United States using indices from racial segregation research. Online news segregation is low, higher than most offline news, lower than interactions with neighbors, co-workers or family, and shows no upward trend.
Checked: PDF from web.archive.org, fetched 5 Oct 2026. Page numbers are those of that PDF.
4 findings
Substance p.27: "we expand our sample to include 391 websites for which we have comScore Media Metrix data on average daily visitors, but no Plan Metrix data". Section II.A says Media Metrix data exist for 459 sites and Plan Metrix data for 119, which leaves at most 340 such sites.
- p. 27 Section VI.A, Other Robustness Checks
include391 websites for which we have comScore Media Metrix data on average daily visitors, but no Plan Metrix data
- p. 6 Section II.A
We have at least 1 month of Media Metrix data for 459 of the sites on our list.
- p. 7 Section II.A
Plan Metrix data on ideological composition for 119 of the sites on our list
- p. 7 II.A Data
We have at least 1 month of Plan Metrix data on ideological composition for 119 of the sites on our list.
Only the authors can say what is right here.
Judge's note: I confirmed on p.6 (459 Media Metrix sites), p.7 (119 Plan Metrix sites, available 'only for relatively large sites') and p.27. In Table VIII on p.28 the row is labelled '391 Websites in comScore micro data', which suggests 391 may be the size of the whole expanded set rather than the number of added sites. Nothing in the paper says that any Plan Metrix site lacks Media Metrix data. The counts do not reconcile as written.
Second check: another model agreed
Report a problem with this finding
Reference or label The Figure VIII caption says "highest education level in the household in 2002 comScore micro data (Internet)", but Section II.A lists comScore micro data extracts only for "2004, 2006, 2007, and 2008".
- p. 32 Figure VIII caption, education isolation
highest education level in the household in 2002 comScore micro data (Internet)
- p. 7 Section II.A
We have separate data extracts for 2004, 2006, 2007, and 2008.
Only the authors can say what is right here.
Judge's note: I confirmed on the p.32 image. The same caption uses 2008 comScore micro data for race and income, and p.7 lists no 2002 extract. The year is probably a slip, but only the authors can say which extract was used.
Second check: another model agreed
Report a problem with this finding
Reference or label Reference list gives the Journal of Political Economy volume as 188 for 2010: "Journal of Political Economy, 188 (2010), 355–408." The paper's own volume sequence implies 118.
- p. 40 Reference list, Stromberg and Snyder
Journal of Political Economy, 188 (2010), 355–408.
- p. 40 References, Stromberg and Snyder
Journal of Political Economy, 188 (2010), 355–408
Judge's note: Checked the crop and p.40: the image shows '188 (2010)'. The paper's own reference list gives JPE volume 107 for 1999 (Cutler et al.) and 114 for 2006 (Gentzkow and Shapiro), one volume per year. That puts 2010 at volume 118, so 188 cannot be right and is most likely a digit error. The exact value 118 is inferred from these in-file entries, and it matches what the reviewers recall.
Second check: another model agreed
Report a problem with this finding
Wording The text says "We rescale this measure so the mean across days is equal to the isolation index for our main measure", but the Figure VII caption says "the isolation index is scaled so that its mean across days is equal to 1".
- p. 30 VI.B
We rescale this measure so the mean across days is equal to the isolation index for our main measure.
- p. 31 Figure VII caption
the isolation index is scaled so that its mean across days is equal to 1.
- p. 31 Figure VII caption
the isolation index is scaled so that its mean across days is equal to 1
Judge's note: I confirmed both passages on pp.30–31. The bottom panels of Figure VII are visibly centred on 1, which matches the caption but not the text's description (the main index is 7.5 points).
Second check: another model agreed
Report a problem with this finding
9 minor (typos and the like; most checked by the judge only)
Wording p.17: "Weighting these results by the overall size of the different media shown in Table I, we estimate that the isolation index for all media combined is 5.1 ... removing the Internet would reduce this number to 4.1". A simple share-weighted average of the Table IV indices does not give these values, and the aggregation method is not stated.
- p. 3 Introduction, para on medium comparisons
across all media from 5.1 to 4.1
- p. 17 Text below Table III
Weighting these results by the overall size of the different media shown in Table I, we estimate that the isolation index for all media combined is 5.1.
Only the authors can say what is right here.
Judge's note: I checked Table I (p.14 shares .29/.29/.24/.10/.05/.03), the Table IV indices and the text on p.3 and p.17. A share-weighted average of the per-medium indices gives about 4.1 with the Internet and about 3.7 without it. Under any such average, removing a medium whose index (7.5) is above the average could not lower 5.1 to 4.1. The published figures may come from pooling outlets across media, which adds between-medium sorting, but the paper does not say so. The text gives no other explanation, so I am asking the authors how the combined index was computed. This is not a claim that the figure is wrong.
Second check: another model agreed
Report a problem with this finding
Wording The Introduction says "we use 2008 individual-level data from Mediamark Research and Intelligence", but Section II.B says the data come from "the spring 2007 and spring 2008 waves of the MRI Survey of the American Consumer".
- p. 3 Introduction, first line
we use 2008 individual-level data from Mediamark Research and Intelligence
- p. 7 Section II.B
spring 2007 and spring 2008 waves of the MRI Survey of the American Consumer
Judge's note: Both passages are confirmed on the page images (p.3 and p.7). The introduction's '2008' is shorthand for data that pool two waves. This is a minor imprecision.
Second check: another model agreed
Report a problem with this finding
Wording Table VI, "Share of daily visitors": .687, .147, .094, .065 for ranks 1–10, 11–25, 26–50 and 51+ sum to .993. The note defines each value as "the fraction of total daily unique visitors across all sites".
- p. 24 Table VI, 'Share of daily visitors' column
.687
- p. 24 Table VI, Share of daily visitors column
.687 .147 .094 .065
- p. 23 Figure V, point labelled #10
#10
Only the authors can say what is right here.
Judge's note: I confirmed the values on the p.24 image. The four rank groups look exhaustive, and the sum's shortfall of .007 is larger than rounding of four three-decimal values allows (at most .002). The note does not say whether 'all sites' means a wider universe than the grouped sites. The comparison with Figure V (top 10 at about 0.6, over roughly 459 sites) is not a clean check, because Table VI may cover the 119-site sample. I am asking the authors which entry or which denominator applies.
Second check: another model agreed
Report a problem with this finding
Wording Table VII: "alternet.org .45 .24 .40" and "dailykos.com .45 .24 .40". The two sites have identical values in all three columns.
- p. 26 Table VII, last two rows
alternet.org .45 .24 .40
- p. 26 Table VII, last row
dailykos.com .45 .24 .40
Only the authors can say what is right here.
Judge's note: The p.26 image shows identical rows for alternet.org and dailykos.com. In Table II these sites have different compositions and sizes (16k vs 26k daily UV). No other pair of rows in Table VII matches. This may be a duplicated row, but a coincidence cannot be ruled out from the files, so only the authors can settle it.
Second check: another model agreed
Report a problem with this finding
Wording p.30: "The top two panels of Figure VII show the total number of unique visitors for all news sites". The Figure VII y-axis and caption show 'News Website Visits per Individual' (an average across individuals, roughly 0.02–0.15).
- p. 30 Section VI.B, paragraph 3
The top two panels of Figure VII show the total number of unique visitors for all news sites
- p. 31 Figure VII caption and y-axis
news website visits per individual is the average across individuals of the number of news websites
Judge's note: I compared the p.30 text with the p.31 figure axis and caption. The plotted variable is a per-individual average, not a total count. The time pattern described is unaffected.
Second check: another model agreed
Report a problem with this finding
Typo The Figure VIII caption reads "MRI data (geography other media)". It should read "(geography and other media)", as elsewhere in the same caption.
- p. 32 Figure VIII caption
MRI data (geography other media)
Judge's note: The p.32 image shows the missing 'and' in the education sentence. The other three instances read 'geography and other media'.
Second check: none (judge only)
Report a problem with this finding
Typo The quoted ANES question in Appendix B lists "extremely liberal, liberal, slightly liberal, moderate or middle of the road, slightly conservative, extremely conservative". A plain "conservative" option is absent, unlike the liberal side.
- p. 37 Appendix B, quoted ANES question
slightly conservative, extremely conservative, orhaven’t you thought much about this?
- p. 37 Appendix B
slightlyconservative, extremelyconservative, orhaven’t
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the ANES 7-point item includes a 'conservative' option. The p.37 image shows the asymmetric list: three liberal options but only two conservative ones. The response option was probably dropped from the quotation.
Second check: none (judge only)
Report a problem with this finding
Typo Journal title misspelled in the Mullainathan and Shleifer reference: "American Economics Review, 95 (2005), 1031–1053."
- p. 40 Reference list, Mullainathan and Shleifer
American Economics Review, 95 (2005), 1031–1053.
- p. 40 References, Mullainathan and Shleifer
American Economics Review, 95 (2005), 1031–1053
- p. 40 References, Mullainathan and Shleifer
American Economics Review, 95 (2005)
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the journal's title is 'American Economic Review'. The crop of p.40 shows 'American Economics Review'. Nothing else in the files gives the AER title for comparison, but the Olken entry uses 'American Economic Journal', which follows the same 'Economic' pattern.
Second check: none (judge only)
Report a problem with this finding
Typo Reference list gives a co-author's first name as "Jon": "Lawrence, Eric, Jon Sides, and Henry Farrell." The author is John Sides.
- p. 40
Lawrence, Eric, Jon Sides, and Henry Farrell.
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the co-author of the 2010 Perspectives on Politics article is John Sides. The crop of p.40 shows 'Jon Sides'. The body text (p.~5) uses only the surname, so nothing in the files settles the first name.
Second check: none (judge only)