Economic crisis and benefits of the Internet: differentiated Internet usage by employment status

ABSTRACT

Using data from the Spanish Survey on Equipment and Use of ICTs in Households for 2007–2011, this paper evaluates the effect of employment status on the diffusion of the Internet among the labor force. We use a bivariate probit with sample selection model to account for a potential selection bias that arises because online usage is only observed for Internet users. Our results show that, controlling for income, employment influences online adoption and usage, and we find evidence of a digital divide in adoption and usage by education and age among the labor force. Employed individuals are more likely to have accessed the Internet and used it more frequently than the unemployed and for different activities. However, conditional on adoption, they do not use the Internet for more personal activities. These findings suggest that firms promote and subsidize Internet access, but this sponsored access does not translate into more personal use.

KEYWORDS:

• ICT
• Internet adoption
• Internet usage
• digital divide
• labor market

JEL CLASSIFICATION:

• J20
• J64
• O33

Acknowledgements

We thank two anonymous referees and the editor for constructive comments and suggestions on an earlier version of this paper. We also thank Ricardo Alonso and Rocío Sánchez Mangas for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. As the ICT-H data does not include identifiers that would allow us to track individuals over time, our dataset is not a panel but a pool of cross-sections.

2. Following DAE (2012, 55–118), we classify the activities that individuals are performing online into basic and advanced services according to their level of difficulty. For example, while sending or receiving emails can be considered as a low difficult task, doing an online course might be considered more complex. This varying complexity is also reflected in the level of use of these activities (75% vs. 10%, respectively) and it is usually linked to the level of skills required to perform the activity (DAE 2012).

3. The empirical literature provides a variety of applications of the bivariate probit model with sample selection (see, e.g. Van de Ven and Van Pragg 1981; Boyes, Hoffman, and Low 1989; Lee, Lee, and Eastwood 2003; Goldfarb and Prince 2008; Orviska and Hudson 2009; Grazzi and Vergara 2014). Note that the name for this model is not generally accepted. Van de Ven and Van Pragg (1981) call this model a probit model with sample selection. Amemiya (1985) calls it type 2 Tobit model, while Greene (2009) and Cameron and Trivedi (2009) call this model bivariate probit with sample selection and bivariate sample selection model, respectively. For technical details and more discussion on this model, see Greene (2009) and Cameron and Trivedi (2009).

4. In a random utility framework as described in Greene (2009), represents the strength of individual’s preference for accessing the Internet relative to not going online. The total utility is unobservable, but we observe the choice most preferred by the individual between the two options, that is, the one with the greater utility. Thus, if the utility of accessing the Internet is higher than the utility of not accessing it, the individual will choose to access the Internet, and we observe d_i,1= 1. Similarly, represents the difference between the utility derived from using the Internet and the utility associated with not using it. Conditional on having accessed the Internet, the individual will choose to use the Internet (d_i,2= 1) if this difference is positive.

5. In the survey, questions about online activities are only available for those individuals who have accessed the Internet. Whenever the respondent declares to be engaged in a specific Internet activity, this activity is counted and included as part of the dummy variable ‘High Scope Use'.

6. Internet access estimations are omitted from the table for clarity of presentation. Similar to results reported in Table 3, we find that the probability of access is continuously increasing along the sample period, and it is higher for men, younger, better educated and employed individuals, and that live in urban areas, and working. Internet access estimations are available from the authors upon request.

7. It is worth noting that, unlike other countries, engaging in online job search is not a mandatory requirement to be eligible for unemployment benefits in Spain. Spanish claimants are required to engage in a minimum number of job search actions, although a variety of actions other than online job search count towards this minimum.

8. In the survey, the occupational status of employed persons is grouped into four categories that distinguish between ICT-professionals vs. other occupations and non-manual vs. manual workers. These occupational groups followed the International Standard Classification of Occupations (ISCO) developed by the International Labour Organization. Specifically, the category ‘non-manual workers' consists of individuals in ISCO Major Groups 0 to 5 (such as managers, professional, or technicians), while the category ‘manual worker' comprises Groups 6 to 9 (e.g. craft workers or machine operators). The category ‘ICT-professional' is composed of computing professional and computer associate professionals, while the ‘non-ICT professionals' consist of all other ISCO Minor Groups.

9. The detailed results of this analysis are not included in this section for brevity, but they are available from the authors upon request.

10. To account for endogeneity and properly estimate the model, we consider ‘computer at home' as instrumental variable for having Internet at home as it is correlated with home access but is not a direct determinant of individual online usage.

11. To estimate the three equations accounting for selection and correlation of two online activities, we used two different methods with similar results: (a) the cmp command in Stata (Roodman 2009), and (b) a two-step procedure in which we first estimate the probability of Internet access, calculate the inverse Mill’s ratio, and then use it as an additional regressor in the bivariate probit to jointly estimate the two decision of usage.