WHO CREATES CONTENT?

Abstract

Until the Internet arrived, content creation and distribution was always an expensive, difficult process. With the Internet it is dramatically easier, faster, and cheaper. Some argue that this will move creation out of the hands of elites and lead to wider participation in the public sphere and to enhanced democracy. This paper makes three contributions to this debate. First, it uses a national random sample of the British population. This is much broader than most prior work. Second, it creates the first evidence-based typology of Internet content creation, identifying three types named ‘skilled content’, ‘social and entertainment content’, and ‘political content’. The implicit assumption of many researchers that only one type of content exists is not accurate. Third, using multivariate logistic regression it shows the characteristics of different populations that produce each type of content. Elites have no impact on creation of skilled content. Social and entertainment content is more likely to be created by non-elites. Only creation of political content is significantly and positively associated with elite status. These results clarify inconsistencies in prior studies. Each type of content is produced by a different kind of creator. Thus, type is more than just content; it also describes differences in who creates the content. The varying relationships between elite status and content creation suggest that it is important for future research to pay close attention to the type of content under study when considering possible democratization of creation.

Keywords:

• digital divide
• elites
• mobility
• content creation
• content types
• oxford Internet survey

This article is referred to by:

THE DIGITAL PRODUCTION GAP IN GREAT BRITAIN

COMMENT SOCIAL STRATIFICATION AND CONTENT PRODUCTION

Acknowledgements

I would like to thank Eszter Hargittai and the anonymous referees for helpful comments, and David Sutcliffe for creating Figure 2.

Notes

Hassani (2006) does not study content creation. Her dataset is from the October 2003 Current Population Survey and her dependent variables are participating in online transaction activities and participating in online information search activities.

The SNS variable is created by asking respondents if they have a profile on five SNSs: Facebook, Twitter, Linkedin, ‘dating sites’, or ‘other’. If they answered ‘yes’ to at least one of the five, they were coded as ‘yes’ on the SNS variable.

A preliminary analysis showed that employed and unemployed respondents did not differ in terms of reported content creation.

To check whether Correa's (2010) or Hargittai and Walejko's (2008) results might be influenced by the fact that they study only students, I ran the same PCA on student Internet users only. The results are identical: three components with the same variables loading strongly on each component (table not shown).

BIC also increases in this model, suggesting that one cannot justify the increased model complexity due to the additional variables. The increase in BIC is not too worrisome, however, because it could be fixed by removing variables that were added but turned out to be not significant.

Smith et al. (2009) do not create a multivariate model. The quotes reproduced in this paragraph describe relationships from two-way tables and similar graphics, which is the only analysis that they do. Schlozman et al. (2010) based on the same data also only analyze two-way tables or equivalent graphics.

Although Table 7 contains only respondents who create content, footnote 5 notes that ‘results are robust when the analyses are performed on the entire sample’ (Hargittai & Walejko 2008, p. 253). Logistic regression analyses on the full sample would contrast respondents who post online content versus those who do not, which is exactly what Table 3 does.

Correa's (2010) text on p. 77 says her scale includes 10 items, but the list in footnote 4 on p. 87 contains only nine items.

A puzzle is that a note at the bottom of Table 3 says it is based on Model 6 in the online supplement. Detailed Model 6 results are reported in 10 tables, Table G–Table P in the online supplement (2011b), where the coefficients for high school graduates are significantly different from college graduates for only 1 of the 10 content creation variables (Table P). In a personal communication, Schradie reports that the note is an error and Table 3 is based on Model 5 from the online supplement. Model 6 is more complex than Model 5, adding nine interaction terms. Both Model 5 and Model 6 are based on the combined 17-survey dataset. See the main text for reasons why the results based on the combined dataset may be an artifact.

Schradie (2011a) would like to study ‘digital democracy’ (p. 145) and the ‘digital public sphere’ (p. 165) but, in the reproduced question texts (2011b, Table A), not a single item contains words like ‘politics’, ‘election’, ‘social issues’, ‘democracy’ or ‘synonyms’. No item explicitly measures political content. One message of the present study is that research cannot generalize from variables which are not explicitly political to political content. See Boulianne (2009) for a summary of research on political engagement, which is related to, but not the same as, political content creation.