Skip to main content
ABC News
Conservative, Liberal Sites Both Fueling Race “Conversation”

There are three topics in the American political discourse that trigger something of a Pavlovian response for me: these are Israel, race and media bias. When a story in one of these subject areas creeps to the top of my Twitter feed, I’m pretty much ready to shut the Internet off, go out for a long walk, and call it a day. Subjects like these are prone to hyperbole and hysteria, and it may be next to impossible to have a rational political conversation until the news cycle has run its course.

We pretty much never write about Israel — I mean, literally, we’ve never published an article on it. We do sometimes address media bias — an interesting subject that people say mostly uninteresting things about — but it’s not our main beat. Race, on the other hand, is a subject that’s a little more central to understanding the behavior of the electorate, and so we can’t entirely avoid it.

Nevertheless, the race “conversation” is often counterproductive. It’s not that the issue can’t be talked about intelligently — but as a matter of fact, it usually isn’t; none of us were made any the wiser by the Shirley Sherrod incident, certainly.

I therefore thought it was worth doing some investigation into just who is responsible for driving the obsession with race and racism. Is the new media primarily responsible for propagating the “conversation”, as opposed to traditional news outlets? Are partisan sites more responsible than non-partisan ones?

The basis of my study is a count of how many times various websites mentioned the terms ‘racism’, ‘racial’ or ‘racist’ in the past 30 days, out of the fraction of all articles they published, according to a search on Google News. For instance, 132 of 5752 stories in the Los Angeles Times used one of the three terms above, or 2.3 percent.

A site was included in the analysis if (i) it is currently in the Memeorandum Top 100 — these are the sites that drive the coverage in the political blogosphere; (ii) its contents are contained in Google News; (iii) it had published at least 100 articles in the past 30 days, and (iv) the Google News hits did not consist disproportionately of wire stories which had been republished, which knocked out a few otherwise-eligible websites like the Washington Examiner and Salon. This left me with a sample size of 59.

I then classified the websites according to three criteria:

New Media versus Old Media. A new media website is one which is not associated with a print product, or a radio or television station. An old media website is one which does emanate from a television, radio, newspaper or magazine outlet, and whose content consists primarily of material which is also published in those venues. A hybrid website is one which is technically still associated with a bricks and mortar product like a magazine, but which publishes a significant amount of online-only content or is disproportionately more influential online than in print.

I also developed three subcategories within the old and new media groups, respectively. Old media sites were classified as broadcast (TV/radio), print (newspaper/magazine), or newswires. New media sites were classified into the following groups: ‘neo-traditional’ (these are sites which do a fair amount of original reporting and are essentially online newspapers or magazines, like Huffington Post), ‘watchdog’ (these are sites which focus on media criticism or muckraking), and ‘commentary’.

Obviously, it should be acknowledged that these these distinctions are sometimes blurry. I’m not all that interested in debating how particular sites were classified, and this shouldn’t make much difference given that we have a fairly robust and diverse group of websites.

Politics Focus. Sites which have a primary focus on American politics are marked with an asterisk in the chart you’ll see below. I apply this tag fairly conservatively — for example, and are not classified as having a primarily politics focus. Although MSNBC’s and Fox News’s televised programming probably does have a principally political focus, the content associated with their websites (which is what we’re measuring here) is somewhat more well-rounded.

Liberal, Conservative or Neutral. News outlets which take an explicitly liberal or conservative viewpoint are designated in blue or red, respectively, in the chart below. The key word here is explicitly: I classify a site into one of these groups only if it more or less wears its politics proudly on its sleeve. Neither nor MSNBC get one of these labels — again, I might feel differently if we were evaluating their television programming, but both purport to be neutral and their web content is somewhat more even-handed than their TV channels.

Certainly, some other cases are debatable, and my bias is generally toward classifying a site as neutral rather than placing into one of the partisan camps. A good general test is the volume of headline space that is devoted to storylines which are favorable to liberals or conservatives. For instance, although both Huffington Post and Daily Caller do a fair amount of original reporting, occasionally including stories which are not favorable to their ‘teams’, their frontpages are usually devoted mostly to storylines which tend to be favorable to one particular side. This doesn’t necessarily mean that these sites are biased — which is not at all the same thing — but they have a fairly clear partisan viewpoint.

The results of the study follow. The percentages indicate the proportion of stories that the outlet published in the past 30 days which included the terms ‘racial’, ‘racist’, or ‘racism’.

There are several things to notice here:

— The site which has been most focused on racial storylines, unsurprisingly, is Andrew Brietbart’s Big Government, which played a central role in the Shirley Sherrod controversy. Some 44 percent of stories published at the website in the past month use the terms ‘racial’, ‘racist’ or ‘racism’.

— Some liberal websites also rank very highly. Media Matters mentioned race in 32 percent of its articles, and The Nation in 22 percent of its articles. (, which is arguably Media Matters’ conservative counterpart, wrote about race 21 percent of the time.)

— The figures for websites affiliated with television stations were lower. Stories on mentioned race 3.9 percent of the time, barely any higher than (3.2 percent), (3.0), (3.0) or (1.7), and lower than (4.3 percent). But keep in mind that this reflects content published on the respective websites only. Focusing on their televised content might produce a much different answer; this study does not purport to address that question.

— Outlets which report mostly on domestic politics certainly seem to mention race more often than most newspapers: The Hill does 6.2 percent of the time, the McClatchy DC Bureau 6.5 percent of the time, and Politico 7.5 percent of the time, much higher than other newspapers or newswires. However, these percentages are still fairly low as compared with partisan and/or non-traditional outlets.

— It appears at first that most of the differences are driven by the old media versus new media distinction. Sites which I classify as old media wrote about racism just 2.2 percent of the time, on average, versus an average of 11.3 percent of the time for the hybrid sites and 12.4 percent of the time for the new media sites. But this is potentially a bit misleading: the new media sites also tend to be more explicitly partisan, and tend to have a more explicit and exclusive focus on domestic politics, which also correlate with more frequent mentions of race.

To analyze this problem more robustly, we can do a simple linear regression analysis in which the dependent variable is the percentage of stories which mention racism. (This next bit gets a bit technical.) The independent variables in the regression analysis were a series of four dummy variables: ‘newmedia’ (which takes on a value of -1 for old media sites, 0 for hybrid sites, and 1 for new media sites), ‘focus’ (1 if the site mostly focuses on domestic politics and 0 if it doesn’t), ‘liberal’ (1 if the site has a liberal orientation and 0 if it doesn’t) and ‘conserv’ (1 if the site has a conservative orientation and 0 if it doesn’t).

. regress raceshare newmedia focus liberal conserv

Source | SS df MS Number of obs = 59
-------------+----------------------------- F( 4, 54) = 13.77
Model | .204183574 4 .051045893 Prob > F = 0.0000
Residual | .200186102 54 .00370715 R-squared = 0.5049
-------------+----------------------------- Adj R-squared = 0.4683
Total | .404369676 58 .006971891 Root MSE = .06089

raceshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
newmedia | .0121124 .0134629 0.90 0.372 -.0148791 .0391039
focus | .0397226 .0263053 1.51 0.137 -.0130162 .0924615
liberal | .0473042 .0302723 1.56 0.124 -.0133881 .1079965
conserv | .0921375 .0327596 2.81 0.007 .0264584 .1578166
_cons | .0302334 .0149248 2.03 0.048 .0003109 .0601559

The regression finds that only the ‘conserv’ variable is clearly statistically significant, but ‘focus’ and ‘liberal’ are also promising. On the other hand, whether the site is an old media or new media outlet does not appear to make much difference. If we eliminate the ‘newmedia’ variable and re-run the analysis, we get the following:

. regress raceshare focus liberal conserv

Source | SS df MS Number of obs = 59
-------------+----------------------------- F( 3, 55) = 18.15
Model | .20118289 3 .067060963 Prob > F = 0.0000
Residual | .203186786 55 .003694305 R-squared = 0.4975
-------------+----------------------------- Adj R-squared = 0.4701
Total | .404369676 58 .006971891 Root MSE = .06078

raceshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
focus | .0473919 .0248426 1.91 0.062 -.0023938 .0971777
liberal | .0576054 .0279749 2.06 0.044 .0015425 .1136683
conserv | .1021085 .0307746 3.32 0.002 .0404349 .1637821
_cons | .0206681 .0104562 1.98 0.053 -.0002867 .0416228

Now, the conservative variable is even more clearly statistically significant. Meanwhile, ‘liberal’ has become statistically significant at the 95 percent certainty threshold, and ‘focus’ is very close to it.

Another experiment is to look only at the 25 media outlets that have a primarily political focus. This brings us down to just two independent variables, ‘liberal’ and ‘conserv’:

. regress raceshare liberal conserv if focus==1

Source | SS df MS Number of obs = 25
-------------+----------------------------- F( 2, 22) = 2.53
Model | .042672926 2 .021336463 Prob > F = 0.1030
Residual | .1858856 22 .008449345 R-squared = 0.1867
-------------+----------------------------- Adj R-squared = 0.1128
Total | .228558526 24 .009523272 Root MSE = .09192

raceshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
liberal | .0683176 .0474674 1.44 0.164 -.0301239 .1667591
conserv | .1088037 .0484463 2.25 0.035 .0083323 .2092751
_cons | .0613649 .0375263 1.64 0.116 -.01646 .1391897

The coefficients have not changed much, although the significance values have gone down, partly because we now have a smaller sample size. ‘conserv’ remains statistically significant at the 95 percent threshold, however, while ‘liberal’ is only significant at the 80 percent threshold.

Finally, we can check for robustness by eliminating the website Big Government, which is arguably something of an outlier, from the analysis.

. regress raceshare liberal conser if focus==1 & outlet~="Big Government"

Source | SS df MS Number of obs = 24
-------------+----------------------------- F( 2, 21) = 2.37
Model | .022902641 2 .011451321 Prob > F = 0.1175
Residual | .101254899 21 .004821662 R-squared = 0.1845
-------------+----------------------------- Adj R-squared = 0.1068
Total | .12415754 23 .005398154 Root MSE = .06944

raceshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
liberal | .0683176 .0358577 1.91 0.071 -.0062526 .1428878
conserv | .0745192 .0375009 1.99 0.060 -.0034682 .1525066
_cons | .0613649 .028348 2.16 0.042 .0024119 .1203178

Even with Big Government eliminated, the conservative variable remains statistically significant at the 90 percent threshold. Interestingly, eliminating Big Government also makes ‘liberal’ somewhat more statistically significant, it also once again surpasses the 90 percent confidence level.

None of these model specifications is necessarily better or worse than the others, by the way — it helps to look at all of them in concert. If we do that, we can come to the following conclusions (those of you who are mathophobic are welcome to start reading again now):


— Conservative political websites almost certainly talk about racism more frequently than non-partisan political websites.

— Liberal political websites probably also talk about racism more often than non-partisan political websites.

— We cannot say with a great deal of confidence that conservative political websites talk about racism more frequently than do liberal political websites, although this is more likely than not.

— Sites which concentrate mostly on domestic politics very probably talk about racism more often than those with a broader focus.

— There is not an obvious causal relationship between the business model and the frequency of mentions of race. Although, as a matter of fact, ‘new media’ websites do talk about racism quite a bit more often than old media websites, this is likely driven by factors other than the respective business models themselves. However, a more complete study would need to look at other mediums beyond web publishing, and in particular television.


I don’t necessarily mean to convey the sense that everyone is equally at fault here. Certainly, there are relatively more intelligent ways to talk about racism and relatively less intelligent ones — compare Te-Nehisi Coates against Jeffrey Lord, for instance.

Nevertheless, there seems to be something of an arm’s-race mentality here. Although conservative sites talk about race much more often than the “lamestream media” does, the same is probably also true of liberal sites. Of course, a lot of the time, the sites are talking about racism because they think the mainstream media isn’t talking enough about it: liberals think the media isn’t focused enough on demonstrations of racist behavior at tea parties; conservatives think that there’s a double-standard and that the media isn’t focused enough on the NAACP’s “reverse racism” or an incident or two involving the New Black Panther Party.

I happen to disagree with both of those complaints. Overt displays of racism have been rare at tea-party rallies; to focus on the few exceptions is a form of nutpicking. Moreover, although racism may be a minor ingredient in the tea party “soup”, to posit a central role for it while ignoring the impact of more immediate causes like the economy seems contrived. The media’s coverage of something like the alleged voter intimidation incident in Philadelphia, on the other hand, seems proportionate to its isolated nature.

If it were merely just a matter of people trying to “work the refs”, that would be one thing. But my view is even a bit more cynical than that. It seems to me that much of the “conversation” about race — most of which is taking place between white liberal elites and white conservative elites — really isn’t about at race all. Instead, the issue has simply become another front in the banal, day-to-day struggle for traffic and eyeballs and ratings points, something used on slow news days to give babble an air of gravitas, and name-calling the air of moral authority.

EDIT: The original version of this article misclassified National Review as a liberal media outlet — not an easy mistake to make! Fortunately, this doesn’t really have a tangible impact on our conclusions.

Nate Silver founded and was the editor in chief of FiveThirtyEight.