Abstract
The Internet has the potential to reduce search frictions by allowing individuals to identify faster a larger set of available options that conform to their preferences. One market that stands to benefit from this process is that of marriage. This paper empirically examines the implications of Internet diffusion in the USA since the 1990s on one aspect of this market—marriage rates. Exploring sharp temporal and geographic variation in the pattern of consumer broadband adoption, I find that the latter has significantly contributed to increased marriage rates among 21–30 year-old individuals. A number of tests suggest that this relationship is causal and that it varies across demographic groups potentially facing thinner marriage markets. I also provide some suggestive evidence that Internet has likely crowded out other traditional meeting venues, such as through family and friends.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
This was the first major amendment of the US telecommunications law since the Communications Act of 1934. Among the primary goals of the 1996 Act was the deregulation of the broadcasting market and the promotion of competition in the telecommunications industry by encouraging the entry of any communications business in the market. It was also the first time that the Internet was included in broadcasting and spectrum allotment.
Labor market outcomes that have been considered are unemployment duration, employer-to-employer flows, job search behavior, and the quality of employment matches. See Kuhn and Skuterund 2004; Kuhn and Mansour 2014; Fountain 2005; Stevenson, and Hadass 2004. For other outcomes, also see Brown and Goolsbee 481; Liebowitz and Zentner 2012; Goolsbee and Guryan 2006; and Kroft and Pope 2014.
Rosenfeld and Thomas 2012 conduct a descriptive sociological analysis of the ways the Internet has changed the market for romantic partners in the USA. His analysis is based on data from the How Couples Meet and Stay Together survey (HCMST). It is a nationally representative survey of 4,000 adults of whom almost 80 % are romantically linked to a spouse. Nontraditional, same-sex couples are oversampled. For a similar study covering Australia, the UK, and the USA, see Dutton et al. 2008.
There is an extensive literature linking early marriage for women to lower educational achievement and poverty (Klepinger et al. 1995; Dahl 2010) and early fertility to lower labor market earnings (Blackburn et al. 1993; Loughran and Zissimopoulos 2009). Goldin and Katz (2002) and Bailey (2006) also discuss how another technological phenomenon, oral contraception, affected labor force participation and the occupational choices of women in the 1970s via increases in the age at first marriage and age at first birth.
Median age at first marriage for women (men) was 23.9 (26.1) in 1990, 24.5 (26.9) in 1995, 25.1 (26.8) in 2000, and 25.3 (27.1) in 2005 (census estimates).
In a survey organized by Pew Internet (2005), more than 50 % of the Internet users agreed that misinformation regarding the true marital status is an important concern. Moreover, 20 % of those looking for a partner but who have never used online personals revealed that the lack of trust towards these sites is the main reason why they have not explored this venue. However, despite the fact that some stigma regarding online dating still persists, most Internet users do not view it simply as the last resort.
These arguments, essentially, also imply an ambiguous effect of the Internet on divorce and remarriage. If targeted search leads to matches of more compatible people, such matches will likely be more stable. However, if meeting people becomes easier at all times and ages so that a divorce seems less costly, then this could imply entering a marriage less thoughtfully to begin with. In the latter case, we might expect higher marriage rates (related to Internet expansion) but also higher incidence of divorce. I provide some preliminary suggestive evidence on the link between Internet diffusion and divorce in Section 5.5.
This figure is constructed as follows. Consider, for instance, the cohort aged 21 years old in 1980. This cohort will be 25 years old in 1984, 30 in 1989, 35 in 1994, and 40 in 1999. Hence, I calculate the fraction of ever married population aged (i) 21 in 1980, (ii) 25 in 1984, (iii) 30 in 1989, (iv) 35 in 1994, and (v) 40 in 1999 and then plot these five points on the graph.
(i) High-speed lines are defined as those that provide speeds exceeding 200 kbps in at least one direction. (ii) Prior to June 2005, providers with fewer than 250 high-speed lines in service in a particular state were not required to report data for that state. Small providers of high-speed connections are therefore underrepresented in the earlier data. This change in the filing requirement potentially introduces some measurement error in the measure of broadband diffusion, which likely attenuates my estimates. (iii) The FCC lumps, together in a single category, the high-speed lines that connect residential and small business end-user customers (as opposed to medium and large business, institutional or government end-user customers). A high-speed line is considered as being provided to an end user in the residential and small business category, if the customer orders a high-speed service of a type that is normally associated with residential customers. Unfortunately, there is no way to infer the share of these lines that are exclusively ordered by residential customers (as opposed to small businesses) and so the provided combined measure is the closest information available to household broadband adoption. This remains a limitation of the analysis and likely consists another potential source of measurement error. Henceforth, when the terms residential or home/household broadband/high-speed lines are used in the text to refer to the measure of broadband diffusion followed in this paper, it should be understood that this limitation has been applied. (iv) For confidentiality purposes, the FCC does not report the information on broadband deployment for Wyoming in 2000.
See Forman et al. (2012) for a similar strategy. The March CPS supplements provide annual information on the marital status of the respondents for all years since 1962.
It is worth noting that the broadband’s rate of diffusion has outpaced that of many other popular technologies in the past such as video cassette recorders and personal computers (NTIA 2004).
Most social networking and online dating sites such as Facebook, Google + , Classmates.com, eHarmony, and Yahoo Personals were lunched in the post-broadband era, that is, after 1996.
(i) An additional potential indicator of the surrounding economic environment, which can also influence the decision of marriage, is the evolution of house prices (Lehr et al. 2006). This variable is available from the census for only a subset of years of my baseline sample. For this reason, it is not part of the standard set of covariates used for model estimation. Controlling for this factor, however, in the subset of available sample years, essentially left the estimates unchanged. Furthermore, cellular telephony was another communication technology that began to diffuse before 1996 and which is likely complementary to broadband. This is because Internet connection can also be achieved through cell phones. FCC reports data on mobile phone subscribership by state in the years after 1998. Restricting the sample to the 2001–2006 period and controlling for cell phone adoption does not change the baseline estimates. Results are available upon request. (ii) To define high-tech industries, I follow Hecker (1999).
In an omitted analysis but available upon request, I supplemented the baseline specifications (Table 1) with an additional covariate, measuring the state’s household telephone penetration. The latter might be especially relevant as a potential proxy for the role of dial-up connection, which requires a telephone line. This control was not statistically significant in any specification and its inclusion did not modify any of the baseline estimates.
Individuals who cohabit as unmarried partners can only be identified in the data after 1995. Approximately 5 % of the individuals report living with an unmarried partner in the period 2001–2006. I treat individuals who cohabit as singles. However, the main findings remain intact if I treat them as married or if I eliminate them completely from my sample.
(i) Given the well-documented issues of linear probability models, I have also experimented with a probit specification. The coefficient of Internet diffusion was 0.0059 with a standard error of 0.002. (ii) Rates of 1.2–1.7 % result from 0.0052/0.435 and 0.0075/0.435, respectively, where 0.435 is the percentage of ever married white (non-Hispanic) population aged 21–30 years between 2001 and 2006. (iii) The unit of measurement of broadband diffusion is lines per 100 people in a state. To convert this to units of Internet usage, I proceed as follows. First, I use the 2000, 2001, and 2003 CPS Computer and Internet Use Supplements to construct an individual measure of home Internet usage. Then, I assign broadband penetration by state and survey year and regress Internet usage among 21–30-year-old (non-Hispanic) whites on broadband diffusion, basic individual covariates, state, year fixed effects, and state-specific time trends. The obtained coefficient is 0.012 and statistically significant. This implies that an expansion in household broadband by 1 line per 100 people increases home Internet usage in the focal demographic group by 1.2 percentage points. (iv) The instrumental variable (IV) coefficients are also very similar for men and women (0.013 and 0.01, respectively) and strongly statistically significant. Since both sides of the market are affected by Internet diffusion in very similar ways, in the remaining of the analysis, I pool the two samples together and study the combined effect.
(i) These effects generally remain in the model with state-specific linear time trends. The fraction of the population that is nonwhite as well as state GDP and household income become significant determinants of the decision of females (share nonwhite, GDP) and males (income), respectively, to marry in this extended model. (ii) As marriage and technology diffusion are dynamic processes, it is plausible that a given level of Internet penetration has lagged effects on the marriage response beyond the 3-month lag that is implied by the timing that the two variables are measured. To study the potential presence of such effects, I first use the Akaike criterion to select the model of distributed lags with the best fit. This amounts to estimating (1) with the contemporaneous level of Internet diffusion as well as its third and fifth lag. The estimated coefficients are 0.005 (s.e. 0.0018), -0.0016 (s.e. 0.0033), and 0.0092 (s.e. 0.0062), respectively. While the immediate effect remains almost identical to the coefficient estimated in the column 4 of Table 1, the lagged terms are not individually distinguishable from zero. The three coefficients are, however, jointly statistically significant. Moreover, note that the coefficient of the fifth lag is quantitatively larger than the contemporaneous estimate, implying nonnegligible long-term effects of Internet diffusion: an increase in Internet penetration by one additional line per 100 people in a given year will, all else constant, increase marriage rates by 0.9 percentage points 5 years later. Taking into account the lagged effects, over a 5-year period, a unit increase in Internet diffusion has a cumulative impact on marriage rates of 1.2 percentage points. This suggests an increase in marriage propensity by roughly 2.6 % from an overall mean of 0.484 during the entire period of study (0.012/0.484).
I have also experimented with models including up to a quartic state-specific time trend, with and without allowing for region-year interactions. The results are qualitatively robust and within the range of estimates displayed in Table 2.
As information on the state of residence in the previous year is not reported in 1995, conditioning on cross-state migration results in a small sample size reduction.
I have repeated this falsification test, experimenting with other temporal windows for the dependent variables. I did not find any systematic relationship between future diffusion and past outcomes in any of these cases. This falsification strategy is similar to that of Forman et al. (2012).
I focus on the age group of 25–35-year-old individuals when studying divorce response, because this is when first divorces likely take place. This specification also serves as a useful falsification check for the results presented in Section 5.5 concerning the impact of Internet adoption on other marital outcomes.
Residential telephone adoption is measured by the number of residential phones per family in a given state in 1955. I obtain information on the number of residential telephones in 1955 and the number of families from the 1956 County and City data book. In 1955, there were approximately 0.97 residential telephones per family. State adoption rates, however, varied from 0.40 (Mississippi) to 1.21 (New Jersey). No information is recorded for Alaska, which implies a small reduction in sample size when the IV specification is implemented.
In online Appendix Figs. 2 and 3 , I graph state broadband diffusion in 2005 against that predicted by the share of farm households in 1930 and the share of households in 1960 having hot- and/or cold-piped water (proxy for the presence of basic infrastructure). As expected, the states that adopted advanced telecommunication technologies in recent years are indeed those that had better infrastructure and a smaller share of the population residing in farms. The fact that such characteristics in 1930 and 1960 predict contemporaneous Internet diffusion provides further suggestive evidence in favor of the stability of these factors as determinants of technological diffusion over the course of the last 40 years.
(i) The IV estimates are robust to the exclusion of years 1990–1995 and to the inclusion of regional trends. I have also experimented with the share of farm households by state in 1930 and the share of households in a state having piped water in 1960 as alternative instruments. They are both strong predictors of household broadband adoption and produce very similar effects to the baseline instrument. (ii) The fact that the IV estimate is larger than the OLS is consistent with the presence of measurement error in the baseline regressions. Measurement error is a possibility since prior to 2005, the number of actual high-speed lines was likely underreported, given that smaller providers were not required to report to the FCC (see footnote 9).
One possible way to test the exclusion restriction of the instrument is to check whether telephone penetration in 1955 directly predicts marital patterns prior to the advent of broadband. Under the identifying assumption, the instrument should have no predictable power. Indeed, regressing marriage rates prior to broadband (1985–1995) on the instrument (interacted with year fixed effects) as well as state and year dummies, I find no significant effect between the two quantities.
(i) This is 0.0052* 14/0.548 and 0.013* 14/0.548, respectively, where 0.548 is the percentage of ever married white (non-Hispanic) population aged 21–30 years in 1990. (ii) In light of the calculations explained in footnote 17, an increase in residential broadband penetration by 14 lines per 100 people translates to a sizeable increase in home Internet usage by 17 percentage points. As people use broadband to access the Internet from other locations (such as at work), this is likely only a lower bound of the overall increase in Internet usage induced by broadband.
Table 5 shows OLS and IV estimates in order to provide a lower and an upper bound of the effect.
The March CPS does not formally distinguish between urban and less urban areas. Hence, I define an area as urban on the basis of the metropolitan status of the respondent. I categorize the individual as residing in an urban area if he/she reports living in the central city of a metropolitan statistical area (MSA). If instead a broader definition is chosen whereby urban reflects whether an individual resides in a metro area, then the results clearly indicate that the baseline effects are driven by respondents living in MSAs.
Using the 2005 Pew Internet and American Life Project Survey, I find that conditional on being an Internet user, young African-Americans are as likely as whites to use online personals. This observation pertains to other survey years as well and is also reflective of the fact that, while racial digital divide still persists, the share of African-Americans that are Internet users increased dramatically between 2000 and 2011, especially when compared to other racial groups. The share of Black (non-Hispanic) adults using the Internet increased from 35 % in June 2000 to 71 % in August 2011. For whites (non-Hispanic), the change was relatively more modest climbing from 49 to 80 % within the same period. For Hispanics, Internet usage increased from 40 to 68 % (Pew Internet and American Life Project report; Zickuhr and Smith 2012).
The median time from first marriage to separation is approximately 7 years, and therefore, most transitions from the first marriage to divorce likely take place in the age range of 25–35 years old (Census 2011).
These are nationally representative surveys. In each of the years, a sample was collected of approximately 3,000 people. Across surveys, information only on online dating usage was consistently collected. Hence, no firm conclusion about the effectiveness of other forms of Internet search can be drawn. Individuals also reported information on their state of residence.
The specification is estimated using CPS data for the years 2000 and 2002–2006 for which corresponding information on online dating usage (from the Pew Internet) exists. The model includes the standard covariates employed in all reported specifications. The sample consists of white individuals aged 21–30 years old. The estimate is also robust to the inclusion of region-year interactions. The average (ever) marriage rate among 21–30-year-old individuals between 2000 and 2006 is 0.449. (ii) A model with quadratic state-specific trends produces an even larger effect and also statistically significant. While the possibility of endogeneity of online dating usage cannot be ruled out, the robustness of the estimate to the addition of state-specific trends somewhat alleviates concerns related to omitted variable bias.
I focus on the 2003–2006 period because these are the years for which data are available on both time use and Internet diffusion. ATUS first started to collect time use information in 2003. Nonwork time spent socializing includes socializing, communicating, playing games, doing arts and crafts, listening to music, watching movies, and attending social events with other people (parents, siblings, friends, co-workers, neighbors). Nonwork time spent using the computer includes time spent checking household and personal emails and messages as well as time spent on computer use for leisure (browsing on the Internet, participating in chat rooms, burning CDs, and downloading files, pictures, and music, but excluding playing games).
The specifications also control for the day of the week the survey took place. The estimates remain qualitatively robust to the inclusion of state-specific linear time trends. However, as the sample size is fairly small, the addition of these terms significantly increases standard errors. Furthermore, the results are similar if the analysis includes the entire population of 21–30-year-old individuals and not just singles (never married). Results for time spent socializing are robust to extending the outcome variable to include other social capital-building activities such as volunteering or church attendance.
References
Bailey MJ (2006) More power to the pill: the impact of contraceptive freedom on women’s life cycle labor supply. Q J Econ 121(1):289–320
Becker GS (1973) A theory of marriage I. J Polit Econ 81(4):813–846
Becker GS (1974) A theory of marriage II. J Polit Econ 82(2):S11–S26
Becker GS, Landes EM, Michael RT (1977) An economic analysis of marital instability. J Polit Econ 85(6):1141–1187
Blackburn M, Bloom D, Neumark D (1993) Fertility timing, wages, and human capital. J Popul Econ 6(1):1–30
Blau FD, Kahn LM, Waldfogel J (2000) Understanding young women’s marriage decisions: the role of labor and marriage market conditions. Ind Labor Relat Rev 53 (4):624–647
Brown JR, Goolsbee A (481) Does the internet make markets more competitive? Evidence from the life insurance industry. J Polit Econ 110(3)
Burdett K, Coles MG (1997) Marriage and class. Q J Econ 112(1):141–168
Crandall R, Jackson C (2001) The $500 billion ppportunity: the potential economic benefit of widespread diffusion of broadband internet access. Criterion Economics, LLC
Dahl G (2010) Early teen marriage and future poverty. Demography 47(1):689–718
Dutton WH, Helsper EJ, Whitty MT, Buckwalter G, Lee E (2008) Mate selection in the network society: the role of the Internet in reconfiguring marriages in Australia, the United Kingdom and United States. Social Science Research Network, doi:10.2139/ssrn.1275810
Forman C, Goldfarb A, Greenstein S (2012) The Internet and local wages: a puzzle. Am Econ Rev 102(1):556–575
Fountain C (2005) Finding a job in the internet age. Soc Forces 83(3):1235–1262
Gillett SE, Lehr WH (1999) Availability of broadband Internet access: empirical evidence. Paper presented to the 27th annual telecommunications policy research conference, September 25–27, 1999, Alexandria, VA. http://dspace.mit.edu/bitstream/handle/1721.1/1480/LehrGillettTPRC99_0523.pdf?sequence=1
Goldin C, Katz LF (2002) The power of the pill: oral contraceptives and women’s career and marriage decisions. J Polit Econ 110(4):730–770
Goolsbee A, Guryan J (2006) The impact of Internet subsidies in public schools. Rev Econ Stat 88(2):336–347
Greenstein S, McDevitt RC (2009) The broadband bonus: accounting for broadband internet’s impact on U.S GDP. NBER Working Paper no. 14758. NBER, Cambridge
Hadass YS (2004) The effect of recruiting on the matching of workers and employers. Harvard University, Mimeo
Hecker D (1999) High-technology employment: a broader view. Mon Labor Rev:18–28
Hitsch GJ, Hortaçsu A, Ariely D (2010) Matching and sorting in online dating. Am Econ Rev 100(1):130–163
Hofferth SL, Flood SM, Sobek M (2013) American Time Use Survey Data Extract System: version 2.4 [Machine-readable database]. Maryland Population Research Center, University of Maryland, College Park, Maryland, and Minnesota Population Center. University of Minnesota, Minneapolis, Minnesota
ITU Report (2012) The impact of broadband on the economy: research to date and policy issues. http://www.itu.int/ITU-D/treg/broadband/ITU-BB-Reports_Impact-of-Broadband-on-the-Economy.pdf
King M, Ruggles S, Alexander JT, Flood S, Genadek K, Schroeder MB, Trampe B, Vick R (2010) Integrated public use microdata series, current population survey: version 3.0. [Machine-readable database]. Minnesota Population Center [producer and distributor], Minneapolis, MN
Klepinger D, Lundberg S, Plotnick R (1995) Adolescent fertility and the educational attainment of young women. Fam Plann Perspect 27(1):23–28
Kreider RM, Ellis R (2011) Number, timing and duration of marriages and divorces: 2009. Current population reports. US Census Bureau, Washington, DC, pp 70–125
Kroft K, Pope D (2014) Does online search crowd out traditional search and improve matching efficiency? Evidence from Craigslist. J Labor Econ 32(2):259–303
Kuhn P, Skuterund M (2004) Internet job search and unemployment durations. Am Econ Rev 94(1):218–232
Kuhn P, Mansour H (2014) Is Internet job search still ineffective. Econ J. doi:10.1111/ecoj.12119
Lehr WH, Osorio CA, Gillett SE, Sirbu MA (2006) Measuring broadband’s economic impact. Final Report National Technical Assistance, Training, Research and Evaluation Project # 99-07-13829. http://cfp.mit.edu/publications/CFP_Papers/Measuring_bb_econ_impact-final.pdf
Liebowitz SJ, Zentner A (2012) Clash of the titans: does Internet use reduce television viewing. Rev Econ Stat 94(1):234–245
Loughran DS (2002) The effect of male wage inequality on female age at first marriage. Rev Econ Stat 84(2):237–251
Loughran DS, Zissimopoulos J (2009) Why wait? The effect of marriage and childbearing on the wages of men and women. J Hum Resour 44(2):326–349
Montgomery M, Trussell J (1986) Models of marital status and childbearing. In: Ashenfelter O, Layard R (eds) Handbook of labor economics. Elsevier, pp 205–271
NTIA (2004) A nation online: entering the broadband age. U.S. Department of Commerce, Economics and Statistics Administration. http://www.ntia.doc.gov/report/2004/nation-online-entering-broadband-age
NTIA (2010) Exploring the digital nation: home broadband Internet adoption in the United States. U.S. Department of Commerce, Economics and Statistics Administration. http://www.ntia.doc.gov/report/2010/exploring-digital-nation-home-broadband-internet-adoption-united-states
Prieger JE (2003) The supply side of the digital divide: is there equal availability in the broadband Internet access market?. Econ Inq 41(2):346–363
Rosenfeld MJ, Thomas RJ (2012) Searching for a mate: the rise of the Internet as a social intermediary. Am Sociol Rev 77(4):523–547
Seitz S (2009) Accounting for racial differences in marriage and employment. J Labor Econ 27(3):385–437
Stevenson B (2006) The impact of the internet on worker flows. Working paper. The Wharton School, University of Pennsylvania
Stevenson B, Wolfers J (2007) Marriage and divorce: changes and their driving forces. J Econ Perspect 21(2):27–52
Stevenson B (2009) The Internet and Job Search. In: Autor D (ed) Studies of labor market intermediation. University of Chicago Press, Chicago, pp 67–86
U.S. Department of Commerce, Bureau of the Census. County and city data books: 1952, 1956, 1962, 1967. ICPSR ed. Ann Arbor, MI: Inter-university consortium for political and social research [producer and distributor], 197. doi:10.3886/ICPSR00012.v1
Wolfers J (2006) Did unilateral divorce laws raise divorce rates? A reconciliation and new results. Am Econ Rev 96(5):1802–1820
Zickuhr K, Smith A (2012) Digital differences. Pew internet and american life project, Washington, DC
Acknowledgments
I would like to thank two anonymous referees, Rachana Bhatt, and seminar participants at the University of Montreal and the 2010 CEA meeting for their helpful comments. Funding from the Alexander Onassis Foundation is gratefully acknowledged.
Author information
Authors and Affiliations
Corresponding author
Additional information
Responsible editor: Alessandro Cigno
Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
Bellou, A. The impact of Internet diffusion on marriage rates: evidence from the broadband market. J Popul Econ 28, 265–297 (2015). https://doi.org/10.1007/s00148-014-0527-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00148-014-0527-7