Myers and Hooper (2026) provide a rigorous design to estimate the effect of the iPhone on US county-level fertility rates between 2007 and 2011. This article does not contest their estimates but argues that the underlying causal model is insufficient on three grounds.
Indice
Economic Crisis, Pre-existing Digital Ecosystems, and the Limits of a Monocausal Account of Fertility Decline
Abstract
Myers and Hooper (2026) provide a rigorous design to estimate the effect of the iPhone on US county-level fertility rates between 2007 and 2011. This article does not contest their estimates but argues that the underlying causal model is insufficient on three grounds. First, the socioeconomic profile of treated counties introduces a systematic confound between iPhone access and social class that entropy balancing does not eliminate. Second, 2007 was not year zero of digital socialization: platforms such as BlackPlanet had already shaped differentiated digital ecosystems along ethnic and class lines before the iPhone launched. Third, the temporal coincidence of the iPhone launch with the 2008 recession creates an identification problem that standard econometric controls do not resolve. The null result for Black women finds a more coherent reading in light of all three factors. The article concludes with an epistemological reflection on monocausal identification strategies applied to complex social phenomena.
Keywords: fertility decline, smartphone, economic crisis, digital socialization, community platforms, sociology of technology
1. Introduction
The US general fertility rate held broadly stable between 65 and 70 births per 1,000 women from 1980 through 2007, then began a sustained decline that reached 54 by 2024 – a 22% fall over seventeen years. The economics and demography literature had until recently failed to identify a period-specific cause capable of accounting for this structural break (Kearney et al., 2022).
Against this backdrop, Myers and Hooper (2026) advance the iPhone as the primary causal driver of the decline. Their methodological contribution is genuine: exploiting AT&T’s exclusive carrier agreement with Apple between 2007 and 2011 as a natural experiment, they construct a geographic identification variable that permits causal estimation of iPhone access on county-level birth rates. Placebo tests on Verizon and Sprint produce null results as expected, and estimates are stable across twenty alternative specifications. The identification design is more robust than most of the descriptive literature on the topic.
Yet the robustness of an identification instrument does not guarantee the correctness of the underlying causal model. A natural experiment can reliably estimate a local average treatment effect without the mechanism through which that effect operates being correctly identified. This is the central argument of the present article: Myers and Hooper demonstrate with reasonable credibility that birth rates fell more sharply after 2007 in AT&T-covered counties; they do not demonstrate why, and the policy implications they draw rest on an unidentified mechanism.
The argument proceeds as follows. Section 2 acknowledges the genuine methodological strengths of the paper. Section 3 examines the socioeconomic confound embedded in the research design. Section 4 reconstructs the digital ecosystem that pre-dated 2007. Section 5 analyses the economic crisis as an alternative explanatory channel. Section 6 offers an epistemological reflection. Section 7 proposes an integrative sociological hypothesis. Section 8 draws implications for future research and policy.
2. What the Paper Gets Right (iphone?
Before developing the critique, it is necessary to recognize the genuine methodological value of Myers and Hooper’s contribution. The core strength of the design is that AT&T’s iPhone monopoly between June 2007 and February 2011 is not a construction of the authors but an independent historical fact: Apple’s choice of AT&T as exclusive partner was determined by a five-year commercial contract negotiated for business reasons entirely unrelated to demographic trends. This means that the geographic variation in AT&T coverage on which the entire identification strategy depends is uncorrelated with pre-existing fertility trajectories across counties – the condition that makes a causal interpretation of the observed birth-rate differences possible (Arkhangelsky et al., 2021).
The two estimators address the main confounding problem – that AT&T counties are systematically more urban and more affluent than control counties – through different and complementary mechanisms. The synthetic difference-in-differences (SDID) estimator selects and reweights control counties so that their pre-2007 fertility trajectory runs parallel to that of treated counties, thereby absorbing pre-existing structural differences. The entropy-balanced Poisson event study reweights control counties so that their average demographic characteristics match those of treated counties (Hainmueller, 2012). The two approaches produce consistent estimates, which strengthens the credibility of the result.
Placebo tests further reinforce the causal reading. Applied to counties served by Verizon and Sprint during 2008–2009 – when neither carrier offered the iPhone or Android – the design produces results statistically indistinguishable from zero in the 15–19 and 20–24 age groups. The significant negative result that emerges in 2011 for both carriers, precisely aligned with the gestation-adjusted arrival of Android on Sprint and Verizon, corroborates rather than threatens the smartphone-based interpretation.
In summary: the effect exists. That the iPhone’s introduction had a measurable impact on birth rates in AT&T counties during 2007–2011 is a finding that robustness checks support. What this article questions is not the validity of the local estimate but the correctness of the proposed mechanism and the legitimacy of projecting it onto phenomena of considerably broader scope.
3. The Socioeconomic Confound
The first structural problem concerns the socioeconomic profile of treated units. As Myers and Hooper document in Table 1, counties with AT&T coverage above 90% differ systematically from control counties: the urban population share is 66.4% in treated counties against 28.2% in controls; median household income is $74,000 against $56,300; and the poverty rate is 12.6% against 15.6%.
The entropy balancing procedure corrects these differences but at a substantial cost to sample representativeness. To identify control counties sufficiently similar to treated ones, the method must concentrate all statistical weight on a very small number of units: the effective sample size falls from 1,399 raw controls to only 77. The balancing covers only four aggregate county characteristics and leaves untouched the individual income distribution within counties – the most relevant variable for understanding who could actually afford an iPhone.
This gap is amplified by the device’s cost. At its June 2007 launch, the iPhone retailed at $499 paired with a mandatory two-year AT&T contract at $59.99 per month – a total two-year commitment of approximately $1,940. At a moment of falling median incomes and rising unemployment, this price point selects a specific user profile: young adults with enough disposable income to absorb a significant discretionary expenditure during a recession. This profile coincides almost exactly with the demographic characteristics of treated counties.
The practical risk is that the paper may be measuring not ‘the iPhone reduces births’ but ‘young women with above-median incomes in urban areas reduced births in 2007–2011’, with the iPhone functioning as a proxy for a socioeconomic profile rather than as a cause. The decisive test would require disaggregating the effect by individual income quintile within treated counties. This analysis is absent from the paper.
4. The Pre-existing Digital Ecosystem
Myers and Hooper’s causal model implicitly assumes that 2007 marked a threshold in the history of digital socialization. This assumption is empirically incorrect. At the time of the iPhone’s launch, the United States already hosted a mature and differentiated ecosystem of digital socialization platforms. MySpace (founded 2003) had reached approximately 100 million accounts by 2006. Facebook (2004) had expanded well beyond its original college-only base. MSN Messenger (1999) was the dominant real-time communication tool among young adults. OkCupid (2004) had established itself as a significant online dating platform before the smartphone era (Byrne, 2007).
Crucially, this ecosystem was not demographically uniform. Alongside mainstream networks, a parallel infrastructure of community-specific platforms had developed explicitly for ethnic and cultural communities. BlackPlanet (founded 1999) had reached 15.8 million registered users by January 2007 and was described in the academic literature as a primary digital institution of the African American community (Byrne, 2007). AsianAvenue and MiGente performed analogous functions for Asian American and Latino communities. These were not marginal phenomena: Pew Research Center data show that in 2006 African Americans adopted social networks at 11%, slightly exceeding the 9% rate among White Americans (Pew Research Center, 2015).
From a network sociology perspective, these community-specific platforms represent an infrastructure of ‘bonding’ social capital in Putnam’s (2000) sense: they reinforced ties within existing communities rather than creating new cross-community connections. Their social logic was structurally different from that of Tinder or Facebook Dating – they were not open relational markets but digital community spaces with a strong identity component.
This reconstruction has direct implications for the paper’s null finding for Black women. Myers and Hooper attribute this speculatively to differential iPhone versus Android adoption or to race-specific pre-2007 birth-rate dynamics. A theoretically more coherent explanation emerges from the ecosystem analysis: African American women in AT&T-covered counties already possessed, before the iPhone’s launch, a robust and community-specific digital socialization infrastructure. For this group, the iPhone would not have represented a qualitative shift in the nature of digital socialization but a change in the device through which an already-established practice was conducted. The marginal effect on social behaviour, and consequently on fertility, would therefore structurally be expected to be smaller. This interpretation does not demonstrate the mechanism but provides a theoretically coherent account of an anomaly the paper fails to articulate.
The existence of these pre-existing ecosystems does not demonstrate that 2007 was not a behavioural turning point: it is plausible that the permanent mobility introduced by the smartphone generated additional effects beyond those of desktop-based digital socialization. What the evidence does show is that the implicit ‘year zero’ narrative in the paper is empirically untenable, and that a research design unable to distinguish between these pre-existing ecosystems cannot correctly identify the mechanism through which smartphone access modified fertility behaviour.
5. The Economic Crisis Channel
The iPhone launched on 29 June 2007. The National Bureau of Economic Research dates the onset of the Great Recession to December 2007. The two events are separated by six months. This temporal proximity constitutes a structural identification problem that Myers and Hooper’s econometric controls do not resolve.
The authors include county-level controls for unemployment, poverty rate, median household income, and housing price change. They argue that coefficient stability across four nested specifications constitutes informal evidence that economic conditions are not absorbing the treatment effect. This argument is weaker than it appears. Demonstrating that estimates remain stable when the same economic indicators are added and removed shows only that those particular indicators do not alter the estimates. If the recession operated through channels that none of the included indicators can capture – such as the closure of physical socialization venues or the reduction in household discretionary income – its effect remains invisible across all four specifications, and coefficient stability provides no additional reassurance.
The Great Recession produced a well-documented contraction of civic and community organizations, particularly in higher-income urban areas where voluntary associations, fitness centres, cultural organizations, and paid entertainment venues were concentrated (Pew Research Center, 2015). A plausible channel, not documented in the paper, concerns precisely this physical socialization infrastructure. For urban middle-class young adults, pre-crisis socialization was substantially organized around paid venues: gyms, bars, cultural centres, professional associations, recreational sports leagues. These were not merely entertainment spaces but the institutional contexts in which romantic relationships formed and developed. It is plausible that the recession reduced participation in these venues, causing some to close and altering the cost-benefit calculus of attending those that remained open. The empirical literature on this specific channel is, however, sparse – a limitation explicitly acknowledged in the implications section – and this constraint should be borne in mind when weighing this hypothesis.
The simultaneous increase in social network adoption – from 25% of American adults in 2008 to 50% in 2011 (Pew Research Center, 2015) – is compatible with a substitution effect driven by economic constraint rather than technological displacement: digital socialization expanded not because the iPhone made it attractive but because the recession made physical socialization expensive. The iPhone and the recession are not two independent factors operating in parallel; they are two concurrent shocks interacting through the same behavioural channel.
This alternative channel generates a testable empirical prediction that differs from Myers and Hooper’s iPhone hypothesis. If the iPhone is the operative cause, the effect on birth rates should be concentrated in high AT&T coverage counties regardless of local recession severity. If the crisis channel is operative, the effect should be stronger in counties where the recession produced larger contractions in disposable income and civic infrastructure, independently of AT&T coverage. Myers and Hooper do not perform this interaction analysis – its absence is a gap in the identification strategy the paper does not acknowledge.
6. An Epistemological Note
The limitations identified in the preceding sections reflect a broader epistemological orientation that characterizes a significant strand of contemporary social science research on technology and behaviour: the reduction of complex social causality to single-variable identification strategies, combined with the rhetorical amplification of results beyond what the evidence supports.
This orientation has a recognizable structure. A methodologically sophisticated identification strategy generates a robust estimate of a local average treatment effect over a bounded period and geography. This estimate is then projected onto a phenomenon of considerably larger temporal and social scope – the post-2007 global fertility decline – through a chain of interpretive steps the original design cannot support. The mechanism narrative in Section 8 of Myers and Hooper’s paper is constructed from descriptive national trends explicitly acknowledged as non-causal, yet the conclusions treat them as identified channels.
The epistemological cost of overclaiming is not merely academic. If declining fertility is primarily a consequence of smartphone adoption, the policy implication is technological regulation. If it is primarily a consequence of economic restructuring and the erosion of physical socialization infrastructure, the implication is economic intervention. These are not equivalent prescriptions, and getting the causal story right matters beyond the pages of academic journals.
The alternative proposed here is not causal agnosticism but the adoption of analytical frameworks adequate to the complexity of the phenomenon. Social causality in the domain of fertility behaviour is irreducibly multilevel: it operates simultaneously at the level of macroeconomic structures, institutional configurations, community social capital, individual psychological dispositions, and technological affordances. A research design that can identify only one of these causal levels, however rigorously, cannot claim to have explained the phenomenon – only to have measured one contributing factor under specific conditions.
7. An Integrative Sociological Hypothesis
The three analytical threads developed above do not demonstrate an alternative mechanism to that proposed by Myers and Hooper: they show that the mechanism they hypothesize is not the only one compatible with the data. This section proposes an interpretive framework integrating the three threads into a sociologically coherent reading, with the same epistemological caution required of the authors under critique: what follows is a theoretically motivated hypothesis, not an empirically identified explanation.
The framework can be articulated in terms of the restructuring of the field of romantic socialization in Bourdieu’s (1984) sense. Before the 2008 crisis, romantic socialization for urban middle classes was organized around an ensemble of paid physical institutions that functioned as socially structured encounter spaces, incorporating and reproducing specific courtship habitus, gendered expectations, and hierarchies of social capital.
The 2008 crisis simultaneously eroded two elements of this field: the economic means to frequent paid venues and the very survival of many of those venues. In this context of contraction, the smartphone – and the digital socialization platforms it made mobile and ubiquitous – did not create a social void but filled a void already forming. It did so, however, according to logics structurally different from those of the physical institutions it was replacing: the logics of relational markets, public self-presentation, and algorithmic partner selection (Airoldi, 2022).
This transition from identity-rooted digital communities to digitalized relational markets redistributed trust costs from community institutions to individuals. Research on the perceived risks of online dating platforms documents three categories of cost that fall asymmetrically by gender: identity deception risk, physical and sexual risk, and the psychological risk of emotional exhaustion and systemic distrust (Couch et al., 2012). Women and psychologically more vulnerable individuals bear substantially higher vigilance costs – costs with no equivalent in institutionally structured physical socialization.
From this perspective, the problem is not simply the growth of social networks but the shift from digital communities rooted in pre-existing social memberships to digitalized relational markets in which trust costs are individualized. This process accounts for the decline in births – in particular in unplanned births among younger women (Buckles et al., 2025) – not as a direct effect of the smartphone on sexual behaviour but as an indirect effect of the institutional restructuring of the field in which fertility-generating relationships form.
8. Implications for Research and Policy
The methodological implications concern primarily the adequacy of analytical instruments relative to the complexity of the phenomenon under investigation. Natural experiments based on geographic or temporal discontinuities are powerful tools for estimating local causal effects; they are not adequate tools for explaining long-run demographic phenomena involving multiple interacting causal levels. Combining robust local estimates with explanatory narratives of global scope does not produce stronger knowledge but an illusion of understanding that may be more misleading than the simple correlations it claims to supersede.
For future research, the interpretive framework proposed here suggests at least three empirical agendas. The first concerns the systematic documentation of the collapse of physical socialization infrastructure during the 2008–2011 crisis, disaggregated by income class and geographic area: this literature is surprisingly sparse and its absence is itself a sociologically relevant datum. The second concerns the study of community-differentiated digital ecosystems in the pre-smartphone period, with attention to user distribution by demographic group and to the social logic of each platform. The third concerns the asymmetric costs of the transition from physical to digital socialization by gender, class, and psychological vulnerability.
For policy, the proposed framework suggests that the effectiveness of financial transfers in support of natality may be higher than Myers and Hooper conclude, if the operative mechanism is economic-structural rather than technological. Disposable income determines access to physical and digital socialization institutions, and that access shapes the probability of forming the relationships that produce children. A policy acting on the underlying economic structure is more consistent with the hypothesis advanced here than with Myers and Hooper’s monocausal account.
9. Conclusion
Myers and Hooper (2026) make a genuine methodological contribution to the literature on the US fertility decline. The natural experiment design is solid, the placebo tests hold, and the estimates are stable across alternative specifications. The effect is real: in AT&T-covered counties, birth rates fell more sharply after 2007.
What the design cannot identify is the mechanism through which this effect operates. The paper’s mechanism section is entirely descriptive – as the authors themselves acknowledge – yet the conclusions treat it as if it contained identified causal evidence. This inferential leap is the paper’s central problem.
The three critiques developed in this article do not demonstrate that the iPhone had no effect on fertility, nor that the economic crisis is the true cause of the phenomenon. They demonstrate something more circumscribed and more robust: that the estimated effect is compatible with equally plausible alternative explanations, and that the available evidence does not permit establishing which among these mechanisms is primarily responsible. Some of these alternatives – including the account of the null result for Black women – render the anomaly more coherently than the model the authors propose. In this situation of mechanism under-identification, the policy implications Myers and Hooper draw are premature: not because the effect does not exist, but because we do not yet know through which channel it operates.
The post-2007 fertility decline is real, widespread, and socially consequential. It deserves explanations commensurate with its complexity: explanations that integrate macroeconomic structures, institutional configurations, technological ecosystems, and gender dynamics, rather than reducing to a single technological shock a process that spans nearly two decades of social transformation.
References
Airoldi M (2022) Machine Habitus: Toward a Sociology of Algorithms. Cambridge: Polity Press.
Arkhangelsky D, Athey S, Hirshberg DA, et al. (2021) Synthetic difference-in-differences. American Economic Review 111(12): 4088–4118.
Bourdieu P (1984) Distinction: A Social Critique of the Judgement of Taste. Cambridge, MA: Harvard University Press. (Original work published 1979.)
Buckles K, Guldi M and Schmidt L (2025) The Great Recession’s baby-less recovery: The role of unintended births. Journal of Human Resources 60(1): 224–258.
Byrne DN (2007) Public discourse, community concerns, and civic engagement: Exploring Black social networking traditions on BlackPlanet.com. Journal of Computer-Mediated Communication 13(1): 319–340.
Couch D, Liamputtong P and Pitts M (2012) What are the real and perceived risks and dangers of online dating? Perspectives from online daters. Health, Risk & Society 14(7–8): 697–714.
Hainmueller J (2012) Entropy balancing for causal effects: A multivariate reweighting method to produce balanced samples in observational studies. Political Analysis 20(1): 25–46.
Haidt J (2024) The Anxious Generation: How the Great Rewiring of Childhood Is Causing an Epidemic of Mental Illness. New York: Penguin Press.
Kearney MS, Levine PB and Pardue L (2022) The puzzle of falling US birth rates since the Great Recession. Journal of Economic Perspectives 36(1): 151–176.
King G and Nielsen R (2019) Why propensity scores should not be used for matching. Political Analysis 27(4): 435–454.
Myers CK and Hooper E (2026) Is the iPhone birth control? Causal evidence from AT&T’s 2007–2011 carrier monopoly. NBER Working Paper No. 35310.
Pew Research Center (2015) Social Media Usage: 2005–2015. Available at: https://www.pewresearch.org/internet/2015/10/08/social-networking-usage-2005-2015/ (accessed 22 June 2026).
Putnam RD (2000) Bowling Alone: The Collapse and Revival of American Community. New York: Simon & Schuster.
Twenge JM (2017) iGen: Why Today’s Super-Connected Kids Are Growing Up Less Rebellious, More Tolerant, Less Happy. New York: Atria Books.
