Digital divide and broadband divide some multiple regression results

Size: px

Start display at page:

Download "Digital divide and broadband divide some multiple regression results"

Emerald Turner
5 years ago
Views:

1 Digital divide and broadband divide some multiple regression results Andrew Leigh* Summary: In a Progressive Policy Institute report that I co-authored with Rob Atkinson, Clear Thinking on the Digital Divide (available at we presented multiple regressions from the CPS. The following table carries out these regressions in a slightly better way (probit rather than OLS), and updates them to Since the US Federal Government s digital divide reports have not included multiple regression results, the following table may be a useful supplement to those official reports. Some data on broadband is also presented. Since I have not seen this information elsewhere, it is presented here for the benefit of researchers working on the digital divide who are interested in parsing out the effect of these various factors, but would prefer not to take the time to run these regressions themselves. * PhD student and Inequality Fellow at the Malcolm Wiener Center for Social Policy, Kennedy School of Government, Harvard University. Contact through or andrew_leigh@ksg02.harvard.edu.

2 Digital divide Digital divide and broadband divide some multiple regression results Analysis of the 2001 CPS shows that from 2000 to 2001: The proportion of households with a computer has increased by about 5%, to 56.5%. The fraction with Internet access has risen by around 10%, to 50.6%. (So while over 10% of households had a home computer without Net access in 2001, only 5% were in the same situation a year later.) Both education and income make a respondent s household more likely to tap into new technology. An extra $10,000 raises a household s propensity to own a computer or have Internet access by around 4%, while an extra year of education makes a person s household about 6% more likely to have access to these technologies. Nonwhites have rates of computer ownership and home Internet connection that are around 12% lower than for whites, holding constant income, education, and metropolitan status. Those living outside metropolitan areas have rates that are 5-7% lower than their city cousins. These demographic characteristics (income, education, race and urban status) explain around one-sixth of the variance in computer and Internet usage between households. The remainder is due to all other differences, including regional gaps, number of children, which industry the household members work in, and personal tastes and preferences. I am told by the Bureau of Labor Statistics that no survey of computer use and ownership was conducted in The next survey will be carried out in September 2003, and the results released in late-2003 or early

3 Table 1: What effect do demographics have on households computer and Internet access? Variable Computers Internet Overall 25.47% 36.62% 51.14% 56.56% 19.37% 40.66% 50.64% proportion of households with the technology Income (per extra $10,000) 3.40% (0.0009) 4.19% (0.0009) 4.04% 4.04% (0.0007) 2.45% (0.0007) 3.53% 3.91% (0.0007) Education (per extra year of school) Race (0=nonwhite; 1=white) Locality (0=nonmetro; 1=metro) 4.62% (0.0009) 7.82% (0.005) 3.86% (0.004) 5.72% 9.59% 3.87% 6.40% 12.47% 5.92% 5.95% 12.13% 5.10% 3.31% 6.10% (0.005) 4.27% (0.004) 5.89% 12.44% 6.98% 6.23% 12.47% 6.87% Pseudo R N (households) Data Source: Author s analysis of data from Bureau of the Census, Current Population Survey, Computer Ownership Supplement (1994, 1997, 2000, 2001) Notes: 1 The first number in each cell is the change in the marginal probability correlated with a one-unit increase in the demographic variable, as estimated from a multivariate probit model. Numbers in parentheses are robust standard errors. 2 All coefficients are statistically significant at the 1% level. 3 The dependent variable is whether a household has a computer, or access to the Internet from home. 4 In 1994, respondents were not asked whether they had access to the Internet. 5 Income has been re-coded from standard CPS bands into a scale with an increment of $ Education has been re-coded from CPS bands into number of years of schooling. 7 Race has been re-coded from CPS categories into a white/non-white dummy variable. 3

4 Broadband access in America The 2001 CPS also asked a series of detailed questions about how households connect to the Internet. It found that, of the 50.6% of the households that are connected to the Internet: 80.9% use a regular dial-up telephone line 6.2% use a DSL line 12.1% use a cable modem 0.8% use something else (presumably an Ethernet connection or similar) In other words, it appears that around one-fifth of the American households connected to the Net are using broadband. Broadband divide? What do we know about the households with broadband connections? Again, we can use multiple regression to compare those with dial-up connections to those with broadband. Both education and income make a household more likely to have broadband but only by small amounts. A $10,000 income rise boosts the likelihood by 0.4%, while if the person answering the survey has an extra year of education, the household is 0.8% more likely to have broadband. The starkest difference is between metropolitan and non-metropolitan areas, with city residents being 10% more likely to have broadband reflecting its limited availability in non-urban areas. There is no statistically significant difference between the broadband usage rates of households headed by whites and non-whites. As we would expect with a technology in its infancy, these demographic factors account for only a small fraction of the variance in broadband use just 1.6%. As we have seen with computers and the Internet, this proportion is likely to rise as broadband access rates increase. Table 2 shows the coefficients for this multiple regression. 4

5 Table 2: What effect do demographics have on whether online households connect via broadband? Variable Broadband 2001 Overall proportion of households with 19.09% broadband Income (per extra $10,000) 0.43% Education (per extra year of school) 0.78% Race (0=non-white; 1=white) 1.29% [Not significant] (0.008) Locality (0=non-metro; 1=metro) 10.06% Pseudo R N (households with some form of Internet access) Data Source: Author s analysis of data from Bureau of the Census, Current Population Survey, Computer Ownership Supplement (2001) Notes: 1 The first number in each cell is the change in the marginal probability correlated with a one-unit increase in the demographic variable, as estimated from a multivariate probit model. Numbers in parentheses are robust standard errors. 2 All coefficients except race are significant at the 1% level. Race is not significant at even the 10% level. 3 The dependent variable is whether a household uses broadband to connect to the Internet. The sample includes only households with some form of Internet connection. 4 Income has been re-coded from standard CPS bands into a scale with an increment of $ Education has been re-coded from CPS bands into number of years of schooling. 6 Race has been re-coded from CPS categories into a white/non-white dummy variable. 5

6 What do people use broadband for? There are a number of interesting differences between the computer and Internet usage patterns of those households with broadband, and those without. Table 3 shows some of these patterns. Although these comparisons do not control for demographic factors, multiple regressions (holding constant income, education, race and locality) produce very similar coefficients. As Table 3 indicates, the activities that receive the biggest boost in broadband households are: Employment: working from home is up 29%, looking for a job online is up 13.8% E-learning: the fraction who take a courses online is 67.4% higher for broadband users (though it is still a fairly small proportion of the population) Entertainment: using the Internet to watch TV or movies, or listen to the radio, is up 77.4% This table was included in a Progressive Policy Institute report, released in September 2002: Unleashing the High Speed Internet: Strategies for Stimulating Broadband Demand (available at 6

7 Table 3: How do broadband and dial-up households use computers and the Internet differently? Question (1) Proportion of dial-up households answering yes (2) Proportion of broadband households answering yes Difference (2)-(1) Proportionate difference (2)-(1) (1) This year, have you used a 40.3% 52.0% % computer to work from home? 17.4% 19.8% % Internet to look for a job? 78.5% 79.9% % Internet to complete school assignments? 4.3% 7.2% % Internet to take a course online? 90.6% 92.2% % Internet for or Instant Messaging? 72.3% 78.5% % Internet to get news, weather or sports? Internet for TV, movies, radio? 16.8% 29.8% % Data Source: Author s analysis of data from Bureau of the Census, Current Population Survey, Computer Ownership Supplement (2001). 7

Dissertation Proposal Presentation

Dissertation Proposal Presentation Economic Growth and Welfare Improvement from High-Speed Internet Theeradej Suabtrirat, PhD in Economics Student. Wednesday, May 6 th, 2015 at 2.30-3.30pm. Economics Library