Appendix 3: Additional Results - Other Keyword Categories

Measuring Bias in "Organic" Web Search - Benjamin Edelman and Benjamin Lockwood

To explore whether bias is evident for a wider range of search terms, we analyzed algorithmic results for the 90 most common search terms in the "internet software" category according to Google's Keywords tool. This list includes terms like "ftp server", "spam filter", and "spyware." (As in the body of our paper, we collected all data during August 2010.) We also analyzed algorithmic results for the 10 top search terms from Yahoo Buzz (terms including "iPhone 4" and "Katy Perry").

Within the "internet software" terms, for both Google and Yahoo, odds ratios exceed one in each set of search results, meaning that each search engine tends to favor its own results. However, coefficient estimates for Google were no longer statistically significant. Meanwhile, estimates for Yahoo are highly significant -- indeed, even more so than in the portal service analysis. Within the Yahoo Buzz terms, all three search engines showed signs of bias on first-page results, though only Yahoo's bias was statistically significant.

These results suggest that Google's bias is strongest for the portal service keyword terms discussed in the bodyd of our article. However, Yahoo's bias appears stronger for internet software terms (and perhaps for Yahoo buzz terms, though the odds ratios are undefined for the top 1 and top 3 results, as Google and Bing never place Yahoo's pages among their top results). Moreover, Tables 6 and 8 suggest that Bing's bias appears much stronger when rich results are included in the analysis. (For a full discussion of rich results and their impact, see appendix 4.)

 

Table 5: Probability that a search result references a search engine's own page or service - "internet software" keywords
(odds ratios, p-values in parentheses)
GoogleYahooBing
Top result 1.100 21.118 0.586
(0.78) (0.01) (0.40)
Top 3 results 1.207 2.984 0.865
(0.36) (0.00) (0.68)
First page1.0842.3271.408
(0.50)(0.00)(0.07)
Table 6: Same as above, but includes rich results
(restricted to observations from Google, Yahoo, and Bing)
(odds ratios, p-values in parentheses)
GoogleYahooBing
Top result 0.99 10.65 1.35
(0.99) (0.03) (0.58)
Top 3 results 1.38 2.25 1.25
(0.24) (0.05) (0.55)
First page 1.90 2.506 3.81
(0.00) (0.00) (0.00)
Table 7: Probability that a search result references a search engine's own page or service - top search terms from Yahoo Buzz
(odds ratios, p-values in parentheses)
GoogleYahooBing
Top result 0.737 NA 2.111
(0.73)   (0.41)
Top 3 results 0.809 NA 2.091
(0.68)   (0.19)
First page1.1272.8901.368
(0.65)(0.02)(0.37)
Table 8: Same as above, but includes rich results
(restricted to observations from Google, Yahoo, and Bing)
(odds ratios, p-values in parentheses)
GoogleYahooBing
Top result 7.81 NA 16.15
(0.02)   (0.13)
Top 3 results 3.22 4.857 4.24
(0.02) (0.20) (0.01)
First page 5.50 6.12 2.78
(0.00) (0.00) (0.00)