A Cohort Perspective on Latin America's Fertility Transition

 In the study of Latin America's fertility transition, the sources emphasize a cohort perspective, which tracks the lifetime fertility and socioeconomic outcomes of women born in the same period and place. This methodology contrasts with the more common period approach, which tracks the flow of births at specific points in time. While period measures are often used to understand real-time transitions, the cohort approach is uniquely suited for assessing theories centered on lifetime resources and decisions, such as when to leave school or how many children to have over a life course.

Data Sources and Geographic Focus

The research methodology relies on harmonized census microdata from IPUMS International, utilizing 63 censuses from 17 Spanish- and Portuguese-speaking countries in South, Central, and North America. To allow for more granular analysis, the researchers develop a panel of national and regional birth cohort aggregates. This involves tracking 333 subnational regions (typically states or provinces) to compare how within-region cohort changes in fertility relate to changes in other demographic and socioeconomic variables.

Key Methodological Innovations

• Classification by Place of Birth: A major advantage of using census data in this methodology is the ability to classify women by their place of birth rather than their place of residence. This approach is critical because it rules out potential bias stemming from selective internal migration, where individuals might move to specific areas based on fertility or education decisions.
• Temporal Matching: The cohort approach resolves the "temporal mismatch" found in period data. For instance, it ensures that socioeconomic indicators like child school enrollment rates are linked to the specific cohorts of mothers who actually have children in that age range.
• Sample Restrictions for Accuracy: To ensure the data accurately reflects completed fertility, the researchers focus on women aged 45–49, a period after childbearing is finished but before old-age mortality significantly affects the sample. For child outcomes (schooling and labor), they focus on children aged 12–15 to ensure high rates of maternal coresidence, which allows children to be linked to their mothers’ birth cohorts in the census.

Analytical Framework

The study utilizes a fixed-effect regression framework. This model relates cohort average fertility to other variables—such as mortality, education, and urbanization—while net of region fixed effects and country-by-cohort fixed effects.

• Regional Variation: The identification of trends comes from comparing cohort changes within the same region, rather than making cross-country or simple cross-sectional comparisons.
• Descriptive Nature: The authors explicitly state that their results are descriptive rather than causal. Because many factors like education and fertility may be co-determined, the methodology aims to document how these variables co-evolved rather than establishing definitive cause-and-effect relationships.
• Handling Mortality: The methodology includes a specific focus on the relationship between offspring mortality and fertility. It uses both level-on-level and log-log regression models to determine if fertility declines merely offset mortality improvements or if they outpaced them.

Limitations and Considerations

The sources acknowledge that the definition of "urban" and the scale of administrative divisions vary by country, which can limit cross-country interpretability for specific variables like urbanization. Additionally, while census data is expansive, it may lack certain granular details found in other sources, such as the specific age at marriage. Finally, to maintain precision, the methodology discards any regional cohort cells with fewer than 100 observations.

The sources identify several key demographic and socioeconomic drivers of Latin America's fertility transition, emphasizing that while many factors played a role, women's education and industrialization were the most significant predictors of the decline,.

Dominant Drivers: Education and Industrialization

• Women’s Education: This is cited as the most powerful force in the transition. Gains in women's educational attainment account for 39% of the decline in children ever born and 58% of the decline in surviving children,,.
• Husbands’ Education: While less influential than women's education, rising educational levels for husbands accounted for an additional 9–13% of the fertility decline,.
• Structural Transformation: The shift from agricultural to non-agricultural work—industrialization—accounted for approximately 5–6% of the decline,,. This factor is closely linked with the economic value of children, which tends to be higher in agricultural settings.

Mortality and Survival

A critical demographic driver was the decline in offspring mortality. The research indicates that fertility responses essentially offset improvements in child survival:

• One-for-One Offset: As mortality rates fell, the number of "children ever born" fell at a nearly identical rate,.
• Net vs. Gross Fertility: Because parents were adjusting their birth rates to keep up with mortality decline rather than overshooting it, the number of surviving children did not fall as sharply as the total number of births,,. This suggests that parents focused on reaching a target number of surviving children.

Urbanization and Migration

• Urban Living: Increased urbanization was a predictor of lower fertility in simple models,. However, the sources note that once education and industrialization are accounted for, urbanization loses much of its independent explanatory power. This is partly due to the high correlation between living in a city and working in the non-agricultural sector,.
• Migration: While many women lived outside their birth regions, migration levels were relatively stable across cohorts and were not a primary driver of the broader fertility transition,.

Women’s Employment and Marriage

• Labor Force Participation: Surprisingly, while women’s employment quadrupled across the studied cohorts, it had no residual association with fertility decline once other covariates were adjusted,,. This challenges traditional theories that suggest the "opportunity cost" of a woman's time (specifically entering the workforce) is a primary driver of lower fertility,.
• Nuptiality: The share of women who never married rose only slightly (from 12% to 15% across cohorts). While non-marriage is associated with lower fertility, this demographic shift was too small to be a major driver of the continent-wide transition,.

Challenging the "Quantity-Quality" Tradeoff

One of the most unexpected findings in the sources is that fertility decline was not systematically linked to improvements in child outcomes, such as school enrollment, literacy, or primary completion,,. While these outcomes improved across Latin America, the timing and location of these gains did not track with regional fertility declines. This challenges the popular "quantity-quality" theory, which posits that parents choose to have fewer children specifically to invest more in the education and well-being of each child,,.

In the context of Latin America’s fertility transition, the sources identify women’s education and industrialization (the shift to the non-agricultural sector) as the most powerful socioeconomic predictors of declining fertility. While several other factors—such as urbanization and women's employment—initially appear to be strong correlates, their influence often diminishes once researchers adjust for these dominant forces.

The Dominant Predictors: Education and Sectoral Shift

• Women’s Education: This is the single most significant predictor. Gains in women's schooling account for 39% of the decline in "children ever born" and 58% of the decline in surviving children across the studied cohorts. Education is thought to affect fertility through various mechanisms, including the opportunity cost of time, increased autonomy, and shifting attitudes toward family size.
• Husbands’ Education: The rising educational attainment of men also played a role, accounting for approximately 9–13% of the fertility decline.
• Industrialization: The structural transformation of the economy—specifically husbands moving from agricultural to non-agricultural work—explains roughly 5–6% of the decline. This shift is significant because the economic value of child labor is traditionally higher in agricultural settings.

Surprising Null Results and Complex Associations

The sources highlight several findings that challenge traditional economic theories of fertility:

• Women’s Employment: Although the share of women in the labor force quadrupled across the 1920–1970 cohorts, this increase had no residual association with fertility decline once education and other covariates were adjusted. This challenges theories suggesting that the "opportunity cost" of a woman's time in the workforce is a primary driver of the transition.
• Urbanization: While urbanization is a well-known predictor of lower fertility, the sources indicate it loses its independent explanatory power when industrialization (sectoral composition) is included in the model. This suggests that the rise of industrialized cities, rather than just population density, drove the change.
• Multigenerational Living: Increases in maternal coresidence (living with one's own mother) were initially associated with lower fertility. However, further analysis showed this was largely because these women were more likely to be highly educated or never married, rather than the living arrangement itself being a direct driver of lower fertility.

The "Quantity-Quality" Paradox

The sources find a notable lack of evidence for the "quantity-quality" tradeoff at the regional cohort level. While children’s school enrollment, literacy, and primary completion rates improved dramatically across Latin America, these gains did not systematically track with regional fertility declines. This suggests that while fertility was falling and education was rising, the two processes were not as tightly linked in timing and location as theories of parental investment would predict.

Summary of Predictors

Predictor	Contribution to Fertility Decline	Significance
Women's Education	39% (CEB) / 58% (Surviving)	Dominant force
Husbands' Education	9–13%	Secondary force
Non-Agricultural Work	5–6%	Consistent contributor
Women's Employment	None (after adjustment)	Challenges standard theory
Urbanization	None (after adjustment)	Linked to sectoral shift
Never Marriage	< 2%	Small quantitative impact

One of the most surprising findings in the sources is that Latin America's fertility decline was not systematically linked to improvements in child outcomes, such as school enrollment, literacy, or primary completion. While child outcomes improved significantly across the continent, these gains did not track with regional fertility declines over time.

Improvements Across Cohorts

Across the successive cohorts studied (from mothers born in the 1920s to the 1970s), there were dramatic secular improvements in the human capital of children aged 12–15:

• School Enrollment: Rose from 65% to 89%.
• Literacy: Increased from 83% to 94%.
• Primary Completion: Increased from 41% to 68%.
• Child Labor: The prevalence of work fell from 15% to 8%.

The Lack of Systematic Linkage

Despite these broad improvements, the researchers found that within specific subnational regions, the timing and location of fertility decline did not coincide with the timing and location of improvements in child schooling or work. Specifically:

• Children Ever Born: Had no significant association with school enrollment, literacy, or work in the within-region cohort analysis.
• Surviving Children: Similarly failed to predict variation in school enrollment and work.
• Literacy Paradox: In some models, surviving fertility actually showed a positive association with literacy, the opposite of what standard theories would predict.

Challenging the "Quantity-Quality" Trade-off

These findings challenge the well-known "quantity-quality" (Q-Q) theory, which posits that as parents have fewer children, they invest more in the education and well-being of each child.

• Cross-Sectional vs. Cohort Analysis: The sources note that while there is a strong negative association between family size and schooling in cross-sectional data (comparing different regions at one point in time), this relationship disappears in the regional cohort panel. This suggests that the correlation seen in cross-sectional data is likely driven by other regional factors rather than a direct trade-off between fertility and child investment.
• Broader Context: This result aligns with other recent research, such as a study in Brazil showing that twin births (which unexpectedly increase family size) do not necessarily reduce the schooling of siblings. It also echoes findings from sub-Saharan Africa, where fertility decline has been linked to the education of the mother but not to improvements in the education of her children.

In summary, while Latin American children became much more educated as fertility fell, the sources conclude that fertility decline itself was not a primary driver of these educational gains.

The sources provide a rigorous evaluation of established demographic and economic theories by applying a cohort perspective to Latin America's fertility transition. By tracking lifetime outcomes of women born between the 1920s and 1970s, the research identifies which theoretical frameworks are supported by the regional data and which are challenged.

Theories Challenged by the Data

The study finds little evidence for several prominent theories that have long been used to explain fertility decline:

• Women’s Market Work and Opportunity Cost: Theories proposed by researchers like Schultz (1985, 1997, 2007) emphasize that the rising opportunity cost of a woman’s time—driven by entering the labor force—is a primary driver of lower fertility. However, the sources show that while women's employment in Latin America quadrupled, it had no residual association with fertility decline once other factors were adjusted. This suggests that if opportunity costs mattered, they operated on the intensive margin (hours worked) or through wages rather than the extensive margin of simply being employed.
• The "Quantity-Quality" (Q-Q) Trade-off: Established by Becker and Lewis (1973), Willis (1973), and Caldwell (1980), this theory posits that parents choose to have fewer children (quantity) to invest more heavily in the human capital (quality) of each child. The sources find this theory's support "thin" in the Latin American context: fertility decline at the regional level was not systematically linked to improvements in child outcomes like school enrollment or literacy. While both trends occurred, they did not track each other in timing or location within subnational regions.

Theories Supported by the Data

Conversely, the findings provide strong empirical backing for other theoretical frameworks:

• Women’s Education as a Fundamental Determinant: The data strongly support theories emphasizing women's educational attainment as the most critical factor. Gains in education accounted for 39% of the decline in total births and 58% of the decline in surviving children. This confirms one of demography’s most "durable findings" during this historic transition.
• Structural Transformation (Industrialization): Theories linking fertility decline to the shift from agricultural to non-agricultural work are supported. The transition away from agriculture—where children often have higher economic value for labor—accounted for 5–6% of the fertility decline.
• Mortality Replacement vs. "Hoarding": The research evaluates how parents respond to falling child mortality. It finds that fertility fell "one-for-one" with mortality decline, meaning parents reduced births just enough to offset improved survival. This supports theories of replacement (bearing fewer children because more survive) but challenges theories of "hoarding" (bearing extra children as a hedge against future mortality risk), which would have predicted a sharper drop in surviving fertility.

Methodological Contributions to Theory

The authors argue that a cohort lens is superior for theoretical evaluation because many theories of the fertility transition are centered on lifetime resources and decisions. While period-based data is often used for real-time tracking, it can create a "temporal mismatch"—for example, linking current birth rates to current school enrollment rates for children who belong to entirely different maternal cohorts. By resolving this mismatch, the cohort approach provides a more accurate evidentiary base for assessing whether theories of lifetime fertility actually hold true in practice.

Regulating the Digital Frontier: Evidence from Online Adult Content

 The regulatory context for online adult content is part of a broader global trend where governments are increasingly targeting digital platforms to restrict access to content, often motivated by concerns regarding addiction, misinformation, or harms to minors. This larger landscape of online activity regulation includes diverse efforts such as China’s broad censorship regime, targeted bans on specific platforms like TikTok in India or Twitter/X in Brazil, and social media restrictions for minors in Australia.

Within this broader environment, the regulation of online pornography in the United States has recently shifted toward state-level age verification mandates, which represent some of the most aggressive attempts to regulate online content in the country.

The U.S. Regulatory Framework

The current regulatory push began in earnest with Louisiana’s House Bill 142 on January 1, 2023, which established a template subsequently adopted by 24 additional states. Key features of this regulatory model include:

• Verification Requirements: "Commercial entities" that distribute material "harmful to minors" on websites where such content constitutes a "substantial proportion" must implement reasonable age verification methods. These methods typically involve uploading a government-issued ID, providing credit card information, or using biometric tools like facial recognition.
• Enforcement and Penalties: Enforcement primarily occurs through the court system, where state attorneys general or private individuals can sue non-compliant websites. Penalties for failure to comply are substantial, often reaching thousands of dollars per day of violation.
• Legal Precedent: While previous attempts to protect minors from online obscenity were struck down (e.g., Reno v. ACLU in 1997), the legal landscape shifted in June 2025 when the Supreme Court upheld Texas's age verification law in Free Speech Coalition, Inc. v. Paxton, affirming the government's interest in protecting minors.

Regulatory Challenges and "Leakage"

The sources highlight that regulating online activity presents unique challenges not found in offline markets, leading to various forms of "leakage" that can mute a policy's impact:

• Platform Non-compliance: Dominant sites like Pornhub have chosen to block access entirely in many states to lead legal battles and avoid public relations crises. Conversely, other major competitors like XVideos and XNXX, headquartered abroad, have continued to operate without verification requirements, relying on the practical difficulties of cross-border enforcement.
• Technological Circumvention: Digital consumers can often bypass state-level restrictions using Virtual Private Networks (VPNs) to mask their location. The sources find that roughly 30% of pre-restriction browsing time persisted through such circumvention.
• Alternative Paradigms: Due to these challenges, some proponents—including Pornhub’s parent company, Aylo—advocate for "device-based" age verification. In this model, smartphones or computers would collect and transmit verified age information directly to websites, potentially reducing VPN-based evasion, though this would not prevent users from substituting to non-compliant sites.

Ultimately, the effectiveness of these regulations depends heavily on the ease of technological circumvention and the availability of non-compliant substitutes, factors that policymakers must consider when designing digital regulations for any content category.

Platform responses to age verification laws vary significantly, often dictated by a site's market position, geographic location, and history of legal scrutiny. In the broader context of regulating online activity, these diverse responses create a fragmented landscape that significantly affects the efficacy of any given policy.

The sources identify three primary categories of platform response:

1. Total Access Blocking

The most prominent response was led by Pornhub, the world's most visited adult website, which chose to block access entirely for all users (both adults and minors) in most states that enacted these laws.

• Motivations: The sources suggest Pornhub chose this "aggressive" stance to lead the industry's legal battle against the mandates and to avoid high-profile public relations crises, especially given its history of scrutiny regarding content moderation.
• Legal Strategy: By blocking access, the company could more clearly challenge the laws' constitutionality without risking the massive daily fines (up to $5,000–$10,000 per violation) stipulated in state legislation.

2. Active Compliance

Other platforms chose to remain accessible by implementing the "reasonable age verification methods" required by the laws.

• Mechanism: These sites typically utilize third-party verification providers to handle government-issued IDs, credit card data, or biometric age estimation (such as facial recognition).
• Rationale: This approach allows sites to maintain their user base and revenue streams in regulated states while shifting some of the legal and data-privacy risks to specialized verification firms.

3. Strategic Noncompliance

The sources highlight that a major portion of the adult content market—notably XVideos and XNXX, the second and third most visited sites—chose not to implement any verification systems.

• The "Enforcement Gap": These sites often rely on the practical difficulty of cross-border enforcement. For instance, Pornhub’s parent company (Aylo) is based in Canada, while the parent company of XVideos and XNXX (WGCZ Holding) is based in the Czech Republic, creating jurisdictional hurdles for state attorneys general.
• Impact on Regulation: This noncompliance is a primary driver of "leakage." The study found that 49% of pre-law browsing time was spent on websites that never restricted access, meaning nearly half of the regulated activity continued completely unaffected by the new laws.

Implications for Online Regulation

These varied platform responses demonstrate the difficulty of regulating a global digital frontier with local laws. When a dominant, compliant platform like Pornhub exits a local market, users do not necessarily stop the behavior; instead, they often substitute toward noncompliant competitors (accounting for 10% of baseline consumption) or use VPNs to access the blocked sites (accounting for 31% of baseline consumption). Consequently, the effectiveness of digital regulation depends less on the law itself and more on the uniformity of platform compliance and the ease with which users can find substitutes.

In the broader context of regulating online activity, the sources highlight that digital consumers have a unique ability to adapt to restrictions, making the ultimate impact of such policies uncertain. When U.S. states implemented age verification laws for adult websites, user behavioral responses were characterized by four distinct channels: noncompliance, circumvention, substitution, and cessation.

According to the study, for every 100 hours of pornography consumed before the laws took effect, the breakdown of post-law behavior was as follows:

1. Noncompliance (50 Hours)

The largest share of consumption persisted simply because many websites did not implement the required restrictions. Approximately 49% of pre-law browsing time was spent on websites like XVideos and XNXX that chose not to comply with state mandates, allowing users to continue their habits without interruption.

2. VPN-Based Circumvention (30 Hours)

Users frequently bypassed geographic blocks by using Virtual Private Networks (VPNs) to mask their physical location.

• Persistent Access: Roughly 31% of baseline consumption persisted through this method.
• Young Adult Adoption: The sources find that young adults (aged 18–24) engaged in more VPN-based circumvention than older age groups, likely due to higher technological sophistication.

3. Platform Substitution (10 Hours)

When dominant sites like Pornhub blocked access, users often migrated to noncompliant competitors.

• Market Shift: Approximately 10% of baseline consumption was substituted from compliant sites to those that remained open.
• Concentrated Migration: Most of this traffic flowed to the remaining top-tier noncompliant sites rather than "fringe" adult websites.

4. Cessation (10 Hours)

A minority of users stopped visiting adult websites altogether in response to the regulations.

• Overall Impact: The sources estimate that total pornography consumption fell by approximately 10% (or roughly 0.5 minutes per week for the average user).
• Subgroup Differences: Cessation was notably higher in households with children when using desktop computers, which may suggest that the laws were more effective at reducing access for minors or that parents were more likely to stop using the sites on shared devices. Conversely, cessation was lower among young adults compared to those aged 25–44.

Implications for Digital Regulation

These behavioral responses demonstrate that "leakage"—the continuation of targeted activity through alternative means—significantly mutes the impact of online regulations. Across every subgroup studied, total consumption fell by 15% or less, indicating that while access restrictions did reduce overall activity, the majority of pre-existing behavior persisted through technological workarounds or shifting to alternative platforms. This suggests that the effectiveness of digital regulation depends heavily on the cost of circumvention and the availability of non-compliant substitutes.

In the larger context of regulating online activity, the methodology employed in the sources stands out by using high-frequency, individual-level panel data to overcome the limitations of previous studies that relied on aggregate traffic or search trends. This approach allows for a granular decomposition of user behavior—specifically noncompliance, circumvention, substitution, and cessation.

The research methodology can be broken down into three main components:

1. Data Source and Scope

The study utilizes data from Comscore, a media measurement firm, covering a rotating panel of approximately 550,000 U.S. internet users from January 2022 through December 2024.

• Individual-Level Tracking: The data tracks specific "machines" (desktop and mobile devices), recording the exact timestamp, duration, and number of pages for every website visit.
• Stable Geographic Assignment: Crucially, geographic location is assigned based on stable demographic information rather than contemporaneous IP addresses. This allows researchers to observe browsing activity even when a user is using a VPN, a capability missing from methodologies that rely on IP-based tracking like Google Trends.
• Comprehensive Categorization: Researchers tracked activity across more than 200,000 adult websites, manually coding the top 25 sites as "compliant" or "noncompliant" based on whether they implemented state-level restrictions.

2. Empirical Strategy: Stacked Difference-in-Differences

To identify the causal effect of regulation, the authors employ a stacked difference-in-differences (DiD) design.

• Exploiting Staggered Rollouts: The model takes advantage of the fact that age verification laws and subsequent website shutdowns (primarily Pornhub's) occurred at different times across 25 states.
• Event Study Framework: The analysis compares trends in pornography consumption in treated states to control states (those where shutdowns had not yet occurred or never occurred) during a window ranging from 16 weeks before to 8 weeks after a shutdown.
• Fixed Effects and Clustering: The researchers include machine-by-cohort and calendar-week-by-cohort fixed effects to control for individual habits and time-varying shocks. Standard errors are clustered at the state level to ensure statistical robustness.

3. Data Cleaning and Limitations

The methodology includes specific technical choices to ensure data quality and acknowledges inherent limitations:

• Winsorization: To prevent results from being skewed by extreme outliers (unusually long browsing sessions), all session durations were winsorized at the 95th percentile.
• The "Private Browsing" Gap: A noted limitation is that the data does not capture visits made in private browsing modes (e.g., Chrome’s Incognito Mode). However, the authors argue this does not bias their results because private browsing does not circumvent state-level IP blocks.
• Adult-Only Focus: Because the panel consists entirely of adults, the methodology measures how intended users (adults) respond to laws meant to protect minors.

Methodological Advantages over Prior Research

The sources emphasize that this methodology improves upon existing literature in several ways:

• Quantifying Minutes: Unlike Google Trends, which uses normalized search intensity, this study measures the exact number of minutes spent on platforms.
• Substitution Patterns: It can track substitution to the "full set" of alternative websites rather than just a few popular ones.
• Demographic Heterogeneity: The individual-level data allows researchers to see how responses differ by age, gender, and the presence of children in a household.

The sources perform a heterogeneity analysis to understand how different subgroups of users and device types respond to age verification mandates. This analysis is critical because it examines whether the regulations—intended to protect minors—actually affect users differently based on their age, technological literacy, or household environment.

Key findings from the heterogeneity analysis include:

Age and Technological Sophistication

The researchers focus on young adults (aged 18–24) as a proxy for how minors might respond, given that direct data on minors was unavailable.

• Lower Cessation: Young adults exhibited less cessation (stopping usage) compared to the 25–44 age group.
• Higher Circumvention: This group engaged in significantly more VPN-based circumvention, which the authors attribute to their likely higher level of technological sophistication.

Households with Children

To assess the impact on potential minor access, the study compared desktop machines in households with children to those without.

• Increased Effectiveness: Cessation was larger in households with children present than in those without.
• Interpretations: This could suggest the laws were more effective at reducing access for minors on shared family computers, or that parents in these households were more likely to cease usage altogether once the barriers were implemented.

Device Type and Usage Intensity

The analysis also looked at how the platform (mobile vs. desktop) and the user's baseline habits influenced their reaction:

• Mobile vs. Desktop: Mobile devices showed significantly higher baseline usage (18.8 minutes per week) compared to desktops (2.3 minutes per week). However, desktop users showed higher rates of cessation (approximately 13%) compared to mobile users (approximately 4%) [Figure 3, 95].
• Heavy vs. Moderate Users: Users categorized as "Heavy" Pornhub consumers showed greater cessation in percentage terms than "Moderate" users [Figure 3, 95].

Core Takeaway: The "Leakage" Consistency

Despite these variations, a primary takeaway from the heterogeneity analysis is the consistent attenuation of the law's impact across all groups.

• Universal Persistence: In every subgroup studied—including different genders, income levels, and household types—total pornography consumption fell by 15% or less.
• The Power of Substitutes: Regardless of the demographic, the availability of noncompliant sites and the ease of VPN circumvention provided enough "leakage" to ensure that the vast majority of pre-law browsing behavior persisted.

Ultimately, while certain groups (like young adults) are more adept at circumvention, the presence of close substitutes with low circumvention costs effectively muted the policy's impact across the entire user base.

The sources acknowledge several limitations inherent in their analysis of online adult content regulation, primarily stemming from data constraints and the scope of the study. These limitations are critical for understanding how the findings—such as the observed 10% reduction in total consumption—apply to the broader landscape of digital regulation.

Data Representation and Tracking Constraints

The study relies on a panel from Comscore, which presents specific challenges regarding how accurately it reflects the general population:

• Sample Selection: The Comscore panel is a selected sample of internet users and may not be perfectly representative of the U.S. population in terms of demographics and device-type usage.
• Awareness of Tracking: Because panelists are aware they are being tracked, their browsing behavior—specifically for sensitive content like pornography—might differ from that of the average unobserved user.
• Private Browsing Gap: The data does not capture visits made in private browsing modes (e.g., Google Chrome’s Incognito Mode). While the researchers argue this does not bias the results because private browsing cannot bypass geographic IP blocks, it does mean the study understates total consumption both before and after the laws took effect.

The "Minor" Data Gap

Perhaps the most significant limitation given the regulatory intent is the absence of direct data on minors.

• Primary Target Missing: The age verification laws were specifically designed to protect children under 18, yet the Comscore panel consists entirely of adults.
• Indirect Proxies: Researchers had to use young adults (18–24) and households with children as indirect proxies to infer how minors might react to the mandates. Consequently, the study's estimates reflect the impact on adult users rather than the primary population the laws aim to protect.

Geographic and Technical Measurement

The methodology for assigning users to specific states introduced potential for minor measurement errors:

• Market Mapping: Comscore Markets do not align perfectly with state lines. Researchers had to assign "machines" to states based on population majorities within overlapping markets, which could lead to errors in geographic assignment.
• VPN Identification: While the study can observe browsing activity even when a VPN is used (because geographic assignment is based on stable demographic data), it does not have a direct indicator for whether a VPN is active during a specific session. The persistence of activity on blocked sites like Pornhub is used as a proxy for circumvention.

Scope and External Validity

The sources also highlight limitations regarding the scope of their economic conclusions:

• Welfare Effects: The study documents a reduction in consumption but does not assess the welfare effects of these changes on either the consumers or society at large.
• Supply-Side Exclusion: The analysis focuses exclusively on the demand side (user behavior) and does not examine how these regulations affect the production of adult content or the performers involved.
• Generalizability: While the study identifies universal "leakage" channels like substitution and circumvention, the quantitative estimates are local to the adult website market and may vary in other regulated digital sectors like social media or online gambling.

The Investment Data Implications of the AI Transition

 The source material indicates that the current transition into artificial intelligence is characterized by an unprecedented surge in infrastructure investment, which suggests profound long-term implications for GDP growth and financial markets.

Investment Data and Concentration

AI infrastructure investment is highly concentrated among a small number of publicly traded firms. The "five largest U.S. technology firms"—Amazon, Alphabet, Microsoft, Meta, and Oracle—account for the vast majority of this spending.

• Rapid Scaling: Their combined capital expenditure (capex) was approximately $155 billion in 2022 and is forecast to reach $755 billion by 2026.
• Expansion Trends: Beyond these five, other players like the privately held xAI and "neocloud" providers (e.g., CoreWeave, Lambda) added an incremental $38 billion in capex in 2025.
• Comparison to Historical Booms: By 2026, AI investment is projected to account for 2.4% of U.S. GDP and 13.7% of total gross private fixed investment. This surpasses the peak levels of the late-1990s telecommunications investment cycle, which reached roughly 1.5% of GDP.

Economic Implications and GDP Growth

The sources use a "revealed preference" argument, suggesting that this massive investment implies firms anticipate a significant productivity boom to avoid bankruptcy. The authors model this transition through three primary scenarios based on the number of additional "productivity booms" that may occur between 2028 and 2030:

• Moderate Scenario (No additional booms): Implies a cumulative GDP growth of 5 percentage points by 2030, with the AI sector making up 8% of the economy.
• Transformative Scenario (One additional boom): Results in 20% cumulative GDP growth and an AI output share of 19%.
• Singularity Scenario (Two additional booms): Predicts a massive 58 percentage point increase in cumulative GDP growth, with the AI sector accounting for nearly 39% of the economy by 2030.

Over the very long run (simulated to 2050), expected cumulative GDP growth from AI reaches approximately 30% in the moderate scenario and 231% in the singularity scenario.

Trends in Productivity and Risk

The investment data suggests a productivity increase in the AI sector by a factor of roughly 2.7. However, this rapid transition carries significant economic risks:

• Capital Misallocation: If the anticipated productivity gains do not materialize, the current buildout could be "the largest misallocation of capital in history".
• Asset Pricing: The model predicts that this transition will lead to an increase in the risk-free rate by approximately 0.5 percentage points and a rise in the equity premium by approximately 3 percentage points due to the heightened uncertainty surrounding these rare productivity booms.
• Sector Dominance: As the AI sector's share of the economy grows, its rapid productivity gains will increasingly dominate aggregate U.S. GDP growth, which is expected to reach a long-term annual rate of roughly 7%.

The economic modeling framework presented in the sources is a two-sector open-economy model designed to translate current investment data into long-term productivity and GDP forecasts. This framework specifically focuses on "rare productivity booms" as the primary driver of the transition.

Core Structure of the Model

The model divides the economy into two distinct sectors: AI (sector $a$) and non-AI (sector $n$).

• Production Technology: Both sectors utilize a Cobb-Douglas technology, where output ($Y$) is a function of sector-specific productivity ($z$) and capital stock ($K$).
• Asymmetric Shocks: The defining feature of this framework is that only the AI sector is exposed to rare productivity booms. Unlike "rare disaster" models where a shock might destroy both productivity and physical capital, this model assumes a boom raises productivity while leaving the physical capital stock intact.
• Investment Surge: This asymmetry creates a large gap between a firm's current capital and its now-higher optimal capital, which explains the massive surge in investment currently observed in the data.

Methodology: Revealed Preference

A key component of this framework is the use of revealed preference. Rather than assuming a specific future growth rate, the authors "back it out" from the investment data of value-maximizing firms.

• Calibration of Boom Size ($\xi$): The size of the productivity boom is calibrated to match the observed increase in investment from 2024 to 2027. This implies that the initial boom raised AI-sector productivity by a factor of approximately 2.7.
• Uncertainty and Scenarios: To account for future uncertainty, the model uses a Bernoulli process during a "high-probability window" (2028–2030). By setting the probability of a boom at $0.5$ (maximal uncertainty), the framework generates three distinct scenarios: Moderate (no further booms), Transformative (one additional boom), and Singularity (two additional booms).

Macroeconomic and Financial Integration

The framework extends beyond output to analyze the broader implications for financial markets:

• Adjustment Frictions: While a frictionless model would show immediate capital adjustment, the authors incorporate adjustment costs to produce realistic multi-year dynamics, assuming firms close roughly one-third of the gap between actual and optimal capital per year.
• Asset Pricing: Using Epstein-Zin preferences, the model links these productivity booms to changes in interest rates and risk. Because the AI sector's earnings "load" on the boom, its growth increases the equity premium (up to 3 percentage points) and the risk-free rate (approximately 0.5 percentage points) due to heightened uncertainty and higher expected consumption.
• Sovereign Implications: The framework also notes that higher expected GDP growth can lower a country's debt-to-GDP ratio, potentially compressing the sovereign default premium even as the default-free risk-free rate rises.

The sources analyze the AI transition through a specific two-year window of elevated probability (2028–2030), characterized by "maximal uncertainty" where the likelihood of a major productivity breakthrough is modeled as a 50% annual probability. This framework generates three distinct scenarios—Moderate, Transformative, and Singularity—based on the number of additional "productivity booms" that occur during this window.

The Three Productivity Scenarios

Each scenario is defined by how many additional productivity booms (each multiplying AI-sector productivity by a factor of roughly 2.7) are realized during the 2028–2030 period:

• Moderate Scenario (25% probability): Assumes the initial productivity boom observed in current investment data was a one-time event and no further breakthroughs occur during the window.
• Transformative Scenario (50% probability): Assumes one additional boom arrives during the window, compounding the productivity of the AI sector.
• Singularity Scenario (25% probability): Assumes two additional booms occur back-to-back, drastically increasing AI-sector productivity by a factor of roughly 7.2 beyond its initial level.

Economic Implications by 2030

The sources translate these productivity draws into specific macroeconomic outcomes, showing a wide range of potential impacts on the U.S. economy by the end of the transition period:

Scenario	Additional Booms	AI Share of Economy	Cumulative GDP Growth
Moderate	0	8.0%	5.4%
Transformative	1	19.0%	19.7%
Singularity	2	38.7%	58.2%

(Source:)

Larger Context of the AI Transition

• Deviation from Traditional Estimates: Even the Moderate scenario—which many might consider conservative—predicts a 5 percentage point increase in GDP growth, which is an order of magnitude higher than "task-based" estimates from other economists (such as Acemoglu, who forecasts 0.7 percentage points over ten years).
• Investment as a "Revealed Preference": These scenarios are not mere guesses but are "backed out" from the massive capital expenditures of firms like Microsoft, Alphabet, and Amazon. The model argues that for these firms to avoid bankruptcy given their current spending (projected at $755 billion in 2026), they must be operating under the expectation that one of these higher-growth scenarios is possible.
• The "Singularity" as a Benchmark: The Singularity scenario produces growth that rivals or exceeds the most rapid "growth miracles" in history, such as the postwar Japanese or South Korean economies. By 2030, this scenario envisions the AI sector becoming nearly 40% of the entire U.S. economy.
• Long-Run Growth Trajectory: After the 2030 window, the probability of booms is expected to revert to a long-run steady state of roughly 4% per year. This implies that while the most intense period of transition may end in 2030, the AI sector will continue to drive aggregate U.S. GDP growth at an expected rate of roughly 7% annually in the following decades.

The source material indicates that the AI transition is poised to have a profound macroeconomic impact, primarily driven by a massive surge in infrastructure investment that translates into significant aggregate growth and a structural shift in the composition of the U.S. economy.

Investment as a Share of the Macroeconomy

The scale of AI investment is already reaching historically significant levels relative to the broader economy:

• Share of GDP: AI infrastructure capital expenditure (capex) rose from 0.6% of U.S. nominal GDP in 2022 to a projected 2.4% by 2026.
• Share of Investment: As a portion of total U.S. gross private fixed investment, AI infrastructure grew from 3.3% in 2022 to an estimated 13.7% in 2026, potentially reaching 19.2% by 2027.
• Driving Aggregate Output: By the fourth quarter of 2025, AI investment accounted for approximately one-fifth of the 2.2% year-over-year increase in real GDP; the sources note that without this spending, corporate equipment investment would have been negative.

Projected GDP Growth Scenarios (to 2030)

The macroeconomic impact varies drastically across the model's three scenarios, which depend on the number of additional "productivity booms" realized between 2028 and 2030:

• Moderate Scenario: Adds approximately 5.4 percentage points to cumulative GDP growth by 2030.
• Transformative Scenario: Results in a 19.7 percentage point increase in cumulative GDP growth.
• Singularity Scenario: Leads to a massive 58.2 percentage point increase in cumulative GDP growth.

Even the moderate scenario represents a macroeconomic shift an order of magnitude larger than traditional "task-based" economic estimates, which forecast only a 0.7 percentage point increase over ten years.

Sectoral Shift and Output Shares

The transition is characterized by the AI sector becoming a dominant force in the economy. While the non-AI sector is assumed to grow at its historical rate, the AI sector’s share of total output is projected to rise from roughly 3% today to:

• 8.0% in the Moderate scenario.
• 19.0% in the Transformative scenario.
• 38.7% in the Singularity scenario.

Long-Run Growth and Historical Context

In the very long run (simulated to 2050), the compounding effect of ongoing productivity booms suggests an expected long-term annual growth rate of approximately 7%. Expected cumulative GDP growth from AI could reach 30% (Moderate) to 231% (Singularity) by 2050.

The sources place these impacts in a historical context, noting that the five-year productivity gains envisioned in the higher scenarios (2.7x to 19.5x multipliers) far exceed any historical episode of comparable length, including the U.S. IT boom of the late 1990s. Over a 30-year horizon, the AI sector’s projected impact is comparable to the "East Asian growth miracles" of Japan, South Korea, and China.

The sources place the current AI transition within the framework of historical "rare productivity booms" and "general-purpose technologies," suggesting that while the projected scale is unprecedented in its speed, it shares characteristics with previous major economic transformations.

The U.S. Railroad Era (1850–1910)

The sources identify the U.S. railroad era as the "closest historical analogue" to the current AI infrastructure buildout.

• Expansion and Utility: Between 1850 and 1916, railroad track mileage surged from 9,000 to 254,000 miles.
• The Waste Argument: Critics often cite the eventual abandonment of 63% of peak mileage as evidence of waste; however, the sources argue that this occurred decades later due to the rise of the automobile, not because the original productivity gain was a "mirage".
• Growth Impact: Despite the eventual abandonment of physical capital, GDP per capita nearly tripled (a 2.8x multiplier) during the railroad era, demonstrating that infrastructure booms can justify their costs through massive productivity gains.

Comparative Productivity Multipliers

The sources provide a quantitative comparison of historical growth episodes against the three AI scenarios (Moderate, Transformative, and Singularity):

Episode / Scenario	Period	Multiplier
U.S. IT Boom	1995–2005 (10 yrs)	1.5x
East Asian "Miracles"	~25–30 years	8x–13x
AI Moderate	2024–2029 (5 yrs)	2.7x
AI Singularity	2024–2029 (5 yrs)	19.5x

(Source:)

• Speed of Transition: Over a five-year window, even the Moderate AI scenario (2.7x) far exceeds the productivity multiplier of the 1990s IT boom (1.5x).
• Magnitude: The AI sector's expected multiplier over 30 years (7.2x) is comparable to the "East Asian growth miracles" of Japan, South Korea, and China, which produced 8x–13x multipliers over similar horizons.
• Long-Run Comparison: Over multigenerational horizons (50–80 years), the AI sector's expected multipliers (26.8x–194x) would far exceed those of the Industrial Revolutions (1.7x–2.4x).

General-Purpose Technology and Diffusion

The sources view AI as a General-Purpose Technology (GPT), comparing it to electrification.

• Spillover Effects: Like electricity, the AI sector is expected to reshape other sectors through technological "spillovers".
• Time for Diffusion: The sources note that electrification took roughly 40 years to diffuse from the power sector to manufacturing and services, suggesting the long-term impact of AI may continue to grow long after the initial transition window.

Historical Risks of Over-Investment

The sources acknowledge that the "revealed preference" of managers—spending billions based on expected future gains—has historical precedents of collective over-optimism.

• The Fiber-Optic Buildout: The late-1990s fiber-optic boom is cited as a case where firms invested heavily in capacity that subsequent demand did not justify.
• Capital Misallocation: If the productivity gains from AI do not materialize as expected, the sources warn that the current transition could become the "largest misallocation of capital in history".

Cheapflation Cycles: How Upstream Costs Drive Inflation Inequality

 The core mechanism behind cheapflation cycles is a phenomenon known as pass-through in levels. This concept posits that when firms face upstream cost shocks, they tend to adjust their retail prices on a "dollars-and-cents" basis rather than a percentage basis. Consequently, when the cost of raw materials or manufacturing rises, both premium and budget varieties within a specific product category experience similar absolute price increases.

The Logistics of the Mechanism

While the absolute price increase might be identical across varieties, the percentage impact differs significantly:

• Lower-priced products: A fixed absolute increase (e.g., a few cents per ounce of coffee) constitutes a larger percentage change for cheaper varieties.
• Higher-priced products: The same absolute increase results in a smaller percentage change for premium varieties.

Because low-income households disproportionately purchase lower-priced varieties, they face higher inflation rates than high-income households during periods of rising upstream costs.

Cheapflation in the Context of Cycles

This mechanism generates predictable cycles in inflation inequality that fluctuate alongside upstream cost movements:

• Rising Upstream Costs: When commodity or manufacturing prices spike—as seen during the Great Recession (2008–2011) and the post-pandemic period (2021–2023)—the gap in inflation between low- and high-income households widens significantly.
• Falling Upstream Costs: Conversely, when input costs decrease relative to the aggregate price level, the inflation gap narrows or can even become negative.

The sources indicate that this mechanism is highly effective at accounting for observed fluctuations, explaining nearly 60 percent of the variation in the inflation gap between top and bottom income quintiles for food-at-home purchases.

Theoretical Foundation: Shift-Invariance

The sources reconcile this behavior with models of imperfect competition through a property called shift invariance in demand systems. Under shift-invariant demand, firms maintain positive markups but pass through common cost shocks (like commodity price changes) one-for-one in levels to their retail prices. This explains why firms don't simply maintain fixed percentage markups, which would lead to proportional price increases across all varieties.

Limitations of Official Statistics

A critical consequence of this within-category mechanism is that it is largely invisible in official inflation data. Standard statistics, like those from the Bureau of Labor Statistics (BLS), aggregate prices at a category level (e.g., "roasted coffee"), which masks the internal price divergence between cheap and expensive varieties. Research shows that official data can understate differences in inflation experienced by low- and high-income households by 70–90 percent.

Beyond food, there is evidence that this mechanism applies to other consumption categories, including automobiles, apparel, and services, suggesting that cheapflation cycles are a broad feature of the economy.

Inflation inequality refers to the differences in inflation rates faced by households across the income distribution, a phenomenon that is systematically driven by Cheapflation Cycles. These cycles are characterized by fluctuations in the inflation gap between low- and high-income households, primarily triggered by upstream input cost shocks.

The Mechanism of Inequality

The sources identify pass-through in levels as the fundamental driver of this inequality. When upstream costs (like commodity prices) rise, firms tend to increase retail prices by similar absolute amounts (e.g., cents-per-ounce) across all product varieties in a category.

Because low-income households disproportionately purchase lower-priced varieties, these fixed absolute increases result in a larger percentage price change for their typical consumption basket. For example, in 2011, when coffee commodity prices spiked, the lowest unit price products saw a 40% inflation rate, while the highest price varieties rose by only 12%—even though both had nearly identical absolute price increases in cents-per-ounce.

Cyclical Nature of the Inequality Gap

Inflation inequality is not static but fluctuates in a predictable cycle tied to input costs:

• Rising Input Costs: During periods of high food-at-home inflation (e.g., 2008, 2011, and 2021–2023), the inflation gap between the bottom and top income quintiles widens significantly.
• Falling Input Costs: When input costs decrease, the gap narrows and can even become negative, meaning low-income households temporarily face lower inflation than high-income households.

In food-at-home purchases, this "within-category" mechanism accounts for nearly 60% of the variation in the inflation gap between the top and bottom income quintiles over time. When combined with different expenditure shares across product categories, it explains approximately 70% of the variance.

Understatement by Official Statistics

A major finding in the sources is that official statistics largely mask this inequality. The Bureau of Labor Statistics (BLS) aggregates price data at the category level (such as "roasted coffee"), which hides the internal price divergence between budget and premium varieties. Consequently, official data may understate the difference in inflation experienced by low- and high-income groups by 70–90% during periods of rising costs. For the 2021–2023 period, aggregation understated the growth in food prices for the lowest-income quintile relative to the highest by a factor of seven.

Broader Context Beyond Food

This pattern of inflation inequality extends to various other sectors and units:

• Automobiles: Lower-priced car models and those with lower-income buyer bases exhibit higher inflation sensitivity to overall price changes.
• Geography: Low-income cities experience higher and more volatile inflation because they consume lower-priced varieties that are more sensitive to national cost shocks.
• International Trade: Countries that import lower-priced varieties face greater import price inflation when global commodity prices rise.

Ultimately, the sources suggest that these cycles in inequality are a parsimonious explanation for the disproportionate burden of inflation on the poor during economic shocks, often leaving little room for alternative theories like price-gouging or changes in demand elasticity.

Official statistics produced by agencies like the Bureau of Labor Statistics (BLS) significantly mask the inflation inequality generated by cheapflation cycles because they aggregate price data at a level that is too broad to capture within-category shifts,.

The Core Problem: Aggregation Bias

The primary issue with official statistics is that they report inflation rates pooled across various products within a single category, such as "entry-level items" (ELIs),. Because the mechanism of "cheapflation" operates through pass-through in levels—where absolute price increases hit cheaper varieties harder in percentage terms—the resulting price divergence occurs within these categories,.

The sources highlight several critical ways this aggregation fails to represent the economic reality of low-income households:

• Understating Inequality: Official statistics can understate the difference in inflation experienced by low- and high-income households by 70% to 90%,.
• Invisible Fluctuations: Large swings in the inflation gap are often completely invisible in public data. For example, when coffee commodity prices spiked in 2011, disaggregated scanner data showed an 8 percentage point inflation gap between low- and high-income households within the coffee category alone; however, because the BLS tracks "roasted and instant coffee" as a single item, this gap was lost in the official average,.
• Sensitivity to Upstream Costs: Aggregation masks the fact that food-at-home inflation for low-income households is much more sensitive to upstream producer prices. Official data masks nearly 90% of this disproportionate sensitivity for food manufacturing costs and 72% for farm product costs,.

Case Study: 2021–2023 Post-Pandemic Inflation

The discrepancy between official statistics and actual experiences was particularly "dramatic" during the 2021–2023 period of rising input costs,.

• Official View: Price indices built on ELI-aggregated data showed a negligible difference in food price growth (only 0.3 percentage points) between the bottom and top income quintiles.
• Actual View: Disaggregated data revealed that low-income households experienced 2.4 percentage points higher growth in food prices than high-income households.
• Factor of Seven: In this instance, official aggregation understated the differential growth in food prices for the lowest-income quintile relative to the highest by a factor of seven,.

Generalization Beyond Food

This issue is not limited to groceries; the sources provide evidence that official statistics similarly obscure inequality in other major spending categories:

• Automobiles: The BLS tracks new and used vehicles under broad ELI codes (TA011 and TA021), but does not disaggregate by specific makes and models. Consequently, they miss the fact that lower-priced car models are more sensitive to overall vehicle inflation,.
• Geographic and Import Data: Similar aggregation issues lead to a failure to capture how low-income cities or countries importing cheaper varieties are more exposed to national or global cost shocks,.

The sources conclude that official statistics are especially biased precisely when input costs rise sharply—which is when policymakers are most likely to be concerned about the distributional burden of a rising cost of living.

The sources provide evidence that the mechanism of pass-through in levels and the resulting cheapflation cycles are a broad feature of the economy, extending well beyond the primary data on food-at-home.

Evidence in the Automobile Sector

One of the most detailed areas of evidence outside of groceries is the automobile market. Using microdata on household vehicle purchases and manufacturer-suggested retail prices (MSRPs), the research finds:

• Price Sensitivity: Within a specific vehicle make, a model with a 10% lower initial price exhibits year-over-year growth that is 3.9% to 7.0% more sensitive to average price growth.
• Income Base Sensitivity: Models with a 10% lower-income customer base have inflation rates that are 8.6% to 9.1% more sensitive to overall vehicle inflation.
• Economic Impact: This is significant because vehicle purchases constitute a major share of consumer expenditures—roughly 6.8% of spending between 2021 and 2023, nearly as much as food-at-home.

Evidence Across Other Goods and Services

Using the C2ER Cost of Living Index (COLI), which tracks standardized products across roughly 300 urban areas, the sources identify cheapflation patterns in several other sectors:

• Apparel: Patterns of higher inflation for cheaper items were observed in products like boys' and men's denim jeans.
• Services: Cheapflation was detected in service costs, such as barbershop haircuts and washing machine repair calls.
• Housing and Utilities: The data also showed evidence of these cycles in apartment rents and utility bills.
• General Merchandise: Other goods, including generic medicines (aspirin), newspaper subscriptions, and tennis balls, follow the same pricing logic.

Geographic and Metropolitan Evidence

Beyond specific product categories, the sources use Bureau of Labor Statistics (BLS) metropolitan area price indices to show that this mechanism affects entire cities based on their income levels.

• City-Level Volatility: Consumer price inflation in lower-income cities is higher and more volatile than in high-income cities because they consume lower-priced varieties that are more sensitive to national cost shocks.
• Sector Aggregates: This reduced inflation sensitivity in high-income cities holds true across broad aggregates, including food away from home, goods excluding food, and services excluding shelter.

Implications for Generalization

The consistency of these findings across food, automobiles, apparel, and services suggests that cheapflation cycles are not an isolated phenomenon of the grocery aisle. Instead, they represent a systematic cross-sectional variation in how different income groups experience inflation based on the varieties they consume. The sources conclude that because official statistics aggregate data at the category level (e.g., "new cars and trucks" or "apparel"), they likely miss quantitatively important differences in the cost of living for households across the income distribution in all these sectors.

The theoretical framework of cheapflation cycles is centered on the principle of pass-through in levels, which posits that firms respond to upstream cost shocks by adjusting retail prices on a fixed absolute basis (cents-per-ounce) rather than a fixed percentage basis. The sources provide a rigorous theoretical foundation for this behavior and subject it to extensive robustness testing to ensure its validity across various economic conditions.

Theoretical Foundation: Shift-Invariance

To reconcile the observed "dollars-and-cents" price adjustments with economic theory, the sources highlight a property of demand systems called shift invariance.

• Contrast with Standard Models: In many standard models of imperfect competition, such as those using Constant Elasticity of Substitution (CES) demand, firms maintain fixed percentage markups. This would imply that pass-through in levels is strictly greater than one, as a cost increase would be multiplied by a gross markup.
• Unitary Pass-Through: Shift-invariant demand systems allow for positive markups and downward-sloping demand curves while predicting that common cost shocks are passed through one-for-one in levels. This theoretical property is essential for explaining why premium and budget brands raise prices by the same absolute amount despite having different initial price points.

Robustness of the Mechanism

The sources conduct several tests to confirm that the pass-through in levels is a robust phenomenon and not an artifact of specific data or time periods:

• Exogenous Cost Shocks: To address concerns about reverse causality (e.g., demand shocks affecting commodity prices), the author uses instrumental variables (IV). Specifically, they use exchange rates for major exporters (Brazil and Colombia) and weather shocks to coffee-growing regions to isolate exogenous changes in input costs. The results consistently show uniform pass-through in levels across all price groups.
• Time Horizons: The baseline results use a 6-quarter horizon to measure long-run pass-through. Robustness checks show that changing this horizon to 4 or 8 quarters does not meaningfully alter the finding of uniform pass-through.
• Substitution Effects: A common critique of inflation measurement is the failure to account for consumers switching to cheaper products. The sources test this by comparing the baseline Laspeyres index to a Törnqvist price index, which accounts for substitution. They find that while the absolute level of inflation changes, the inflation gap between high- and low-income groups remains remarkably similar.

Nonhomotheticities and Trading Down

The research further explores how nonhomothetic preferences—where consumption patterns change as real income changes—affect cheapflation.

• Amplification of Inequality: During periods where food prices rise faster than income, households tend to "trade down" to lower-priced varieties.
• Theoretical Irony: Because lower-priced varieties are precisely the products experiencing the highest inflation due to pass-through in levels, the act of trading down actually amplifies the rise in the cost of living for low-income households. Accounting for these nonhomotheticities shows that low-income households face even higher inflation volatility than standard measures suggest.

Quantitative Accuracy

The sources propose a specific mathematical formula (Proposition 1) to predict inflation differences based on the "price gap" between varieties and the "excess inflation" of a category relative to wages. When tested against 18 years of food-at-home data, this theoretical prediction accounts for 70 percent of the observed variance in inflation inequality. Notably, the sources find that the "cheapflation" observed during the 2021–2023 surge was actually slightly less than expected given the massive rise in input costs, suggesting that pass-through in levels is a highly conservative and powerful explanation for inflation inequality.

Labor Market Impacts of Trump 2.0 Immigration Enforcement

 The provided paper, "Labor Market Impacts of ICE Activity in Trump 2.0," by Elizabeth Cox and Chloe N. East, offers the first national, causal empirical analysis of how heightened immigration enforcement during the second Trump administration has affected the U.S. labor market. The study evaluates both the direct impact on undocumented immigrants and the spillover effects on U.S.-born workers, testing the common policy justification that removing immigrants expands job opportunities for native-born citizens.

Methodological Approach

The researchers utilized a natural experiment framework to isolate the effects of enforcement surges.

• Identification Strategy: While enforcement increased nationwide, it did so unevenly across different regions. The authors classified "treated" areas as those experiencing a sudden, large increase in ICE arrests (roughly a doubling) between January and October 2025, while "control" areas did not see such surges.
• Data Sources: The study combined administrative arrest data from the Deportation Data Project with worker-level labor market data from the Current Population Survey (CPS).
• Target Population: The analysis focused on "likely undocumented immigrants" (foreign-born, aged 20-64, with a high school degree or less) and U.S.-born workers in sectors where undocumented labor is over-represented, such as agriculture, construction, manufacturing, and wholesale.

Impacts on Likely Undocumented Immigrants

The study identifies a significant "chilling effect" among immigrants who remain in the U.S..

• Reduction in Work: Even if not physically removed, immigrants in treated areas reduced their work activity out of fear that being at a workplace could lead to ICE interactions. This led to a 4% reduction in employment among the likely undocumented sample.
• Gender Concentration: These effects were primarily driven by males, who accounted for over 85-90% of ICE arrests. Male employment in these sectors dropped by 4.6 percentage points (5%), and their weekly hours worked also fell by 5%.
• Comparison to Past Eras: The chilling effect in Trump 2.0 is notably larger than in previous administrations. The authors calculate that for every one ICE arrest, six male likely undocumented workers stop working. This is more than double the chilling effect observed during the Obama administration, which the authors attribute to the more indiscriminate nature of enforcement in Trump 2.0.

Impacts on U.S.-Born Workers

A central finding of the research is the lack of evidence that U.S.-born workers benefit from these enforcement surges.

• No Substitution: The authors rule out any increase in the U.S.-born employment rate in impacted sectors of more than 0.1 percentage points.
• Complementarity and Harm: Instead of taking the jobs left behind, certain U.S.-born groups were actually harmed. Less-educated U.S.-born men in likely affected sectors saw a 1.3% reduction in employment.
• Economic Mechanism: This suggests that undocumented and U.S.-born workers are complements rather than substitutes. When the supply of undocumented labor is disrupted, it appears to contract overall labor demand in those sectors rather than reallocate it to native workers. Furthermore, there was no evidence that employers raised wages to attract U.S.-born workers to fill vacancies.

Broader Context and Contribution

The study contributes to a growing body of literature on Trump 2.0 by linking national enforcement intensity directly to worker-level outcomes. It concludes that the policy of mass deportation and heightened interior enforcement may have the unintended consequence of harming the very U.S.-born workers it is intended to help by shrinking the labor market in key industrial sectors.

The sources describe the impact on likely undocumented immigrants during the second Trump administration as a significant "chilling effect" that reduces labor supply even among those who are not physically deported. This research provides the first national, causal empirical analysis of these labor market shifts, utilizing a natural experiment framework that compares geographic areas with sudden surges in ICE arrests to those without.

The "Chilling Effect" Mechanism

The study distinguishes between two channels of impact: physical removals and chilling effects. The latter occurs when heightened ICE activity causes individuals to remain in the U.S. but reduce their participation in regular activities—including going to work—out of fear that the workplace could be a site for ICE interactions.

Quantitative Impacts on Employment

The sources highlight several key findings regarding the scale of this impact:

• Overall Employment Reduction: Among likely undocumented immigrants remaining in the U.S. who work in affected sectors, there was a 4% reduction in employment.
• Gender Concentration: These effects are almost entirely driven by males, who represent over 85-90% of those arrested by ICE during the study period. For likely undocumented men, employment dropped by 5% (4.6 percentage points), and their weekly hours worked also fell by 5% (2 hours per week).
• Sector-Specific Impact: The strongest negative effects were observed in the construction sector, which has a high concentration of undocumented labor (over 15%).
• Impact on Wages: The study found no evidence of a consistent effect on pay (hourly wages or weekly earnings) for likely undocumented immigrants in high-impact sectors.

Magnitude and Historical Comparison

The sources quantify the intensity of the Trump 2.0 enforcement regime by comparing it to previous eras:

• The 6:1 Ratio: The authors calculate that for every one ICE arrest, six male likely undocumented workers stop working.
• Trump 2.0 vs. Obama Era: This chilling effect is more than double what was observed during the Obama administration, where roughly 2.3 workers stopped working per detention. The authors attribute this increase to the more indiscriminate nature of enforcement in Trump 2.0 compared to previous administrations.

Broader Labor Market Context

In the larger context of "Trump 2.0," these impacts do not lead to the intended policy goal of creating opportunities for native workers. Instead of U.S.-born workers substituting for the lost immigrant labor, the sources find that undocumented and U.S.-born workers are complements. When the supply of undocumented labor is disrupted, it leads to a contraction of overall labor demand in affected sectors, which ultimately harms less-educated U.S.-born male workers as well.

The study finds no evidence that heightened immigration enforcement in the "Trump 2.0" era expands job opportunities for U.S.-born workers. While a central policy justification for mass deportation is that it creates vacancies for native-born citizens, the researchers found a null effect on the full sample of U.S.-born individuals. Specifically, they were able to rule out any increase in the U.S.-born employment rate of more than 0.1 percentage points in sectors where undocumented labor is prevalent.

Negative Spillovers for Vulnerable Workers

Instead of benefiting from enforcement surges, certain groups of native workers were actually harmed. The sources highlight several key impacts:

• Impact on Less-Educated Men: The most significant negative effect was observed among U.S.-born male workers with a high school degree or less who work in immigrant-heavy sectors. This group saw a 1.3% reduction in their employment rate.
• Sector-Level Correlation: A clear relationship exists across industries: sectors that experienced the largest reductions in undocumented male employment (such as construction) also saw the most significant negative spillover effects for U.S.-born male workers.
• Quantified Loss: The authors calculate that for every six male likely undocumented workers who stop working, one male U.S.-born worker also loses their job.

The Mechanism of "Complementarity"

The researchers explain these negative outcomes through the economic concept of complementarity.

• Jobs are Linked: Rather than competing for the same roles, undocumented and U.S.-born workers often perform different, mutually dependent tasks within the same sector. For example, in construction, while both groups may work as laborers, the overall composition and concentration of their roles differ.
• Contraction of Demand: When the supply of undocumented labor is disrupted by ICE activity, it appears to contract overall labor demand in affected industries rather than reallocating existing jobs to native workers.
• No Wage Growth: The study found no evidence that employers responded to the loss of immigrant labor by increasing wages to attract U.S.-born workers to fill vacancies.

Broader Policy Implications

The findings suggest that the immigration enforcement regime in Trump 2.0 is more indiscriminate than in previous administrations, which amplifies these negative labor market consequences. By documenting these harmful spillovers, the sources conclude that heightened enforcement may ultimately undermine the economic prospects of the very U.S.-born workers it is intended to help.

In the context of the second Trump administration ("Trump 2.0"), the sources characterize the magnitude of immigration enforcement as a historic surge that is significantly larger and more indiscriminate than in recent decades. This intensified activity is measured through sudden, localized spikes in ICE arrests, which the authors use to quantify both the scale of the enforcement itself and its disproportionate impact on the labor supply.

Quantifying the Enforcement Surge

The research identifies a massive national increase in enforcement activity, though the intensity varied geographically.

• Arrest Volume: Following the second inauguration, total daily arrests surged from a baseline of approximately 300 to peaks exceeding 1,100 per day by late 2025.
• Treated Area Growth: Areas classified as "treated" experienced a sudden increase of roughly 200 daily arrests per month relative to control areas. This represented a 114% increase in arrests compared to the pre-period mean.
• Normalized Intensity: Adjusted for population, treated areas saw an increase of 94 arrests per 100,000 non-citizens.
• Gender Concentration: The magnitude of this surge was heavily concentrated among men, with 85% of the increase in total ICE arrests being driven by the arrest of males.

Comparative Magnitudes: Trump 2.0 vs. Prior Eras

The sources emphasize that the current enforcement regime is "much larger and more indiscriminate" than those of the 1930s or the 2010s.

• The Chilling Effect Multiplier: The magnitude of the "chilling effect"—where workers stop working due to fear of ICE interactions—is significantly higher in Trump 2.0 than during the Obama administration.
• Efficiency of Disruption: During the first Obama term, researchers found that for every person detained, roughly 2.3 likely undocumented workers stopped working. In Trump 2.0, this ratio has more than doubled: for every one ICE arrest, six male likely undocumented workers stop working.

Magnitude of Labor Market Disruption

The sheer scale of enforcement has led to substantial shifts in the workforce within treated regions.

• Total Worker Loss: The authors calculate that in the average treated area, there are approximately 7,574 fewer male likely undocumented workers in high-impact sectors like construction and agriculture.
• Impact on U.S.-Born Workers: The magnitude of the enforcement surge also creates a negative spillover for native-born citizens. For every six undocumented male workers who leave the labor force due to enforcement activity, one less-educated U.S.-born male worker also loses their job.
• Underestimation: The authors note that these figures likely underestimate the total economic impact, as their study focuses only on the "chilling effect" among those who remain in the U.S. and does not include the direct labor loss from physical removals and deportations.

The research presented in the paper "Labor Market Impacts of ICE Activity in Trump 2.0" leads to several key conclusions that challenge common political justifications for heightened immigration enforcement. Broadly, the authors conclude that the intensified ICE activity under the second Trump administration has contracted the labor market rather than reallocating jobs to native-born citizens.

1. Heightened Enforcement Creates a Massive "Chilling Effect"

A primary conclusion is that enforcement impacts the labor market far beyond physical removals. The study identifies a meaningful chilling effect where likely undocumented immigrants who remain in the U.S. reduce their work participation out of fear.

• Scale of the Effect: Among the likely undocumented sample in affected sectors, there was a significant 4% reduction in employment.
• The 6:1 Multiplier: The authors conclude that for every one ICE arrest, six male likely undocumented workers stop working.
• Comparison to Past Eras: This chilling effect is more than double what was observed during the Obama administration, which the authors attribute to the indiscriminate nature of enforcement in Trump 2.0.

2. No Evidence of Benefits for U.S.-Born Workers

The study directly addresses the policy justification that removing immigrants creates jobs for native-born workers and finds no evidence to support this claim.

• Null Overall Effect: For the full sample of U.S.-born individuals, the researchers found a null effect, ruling out any employment increase of more than 0.1 percentage points.
• No Wage Growth: There is no evidence that employers responded to labor shortages by increasing wages to attract U.S.-born workers.

3. Negative Spillovers and Harm to Vulnerable Native Workers

The authors conclude that instead of helping, the enforcement surge actually harmed certain segments of the U.S.-born population.

• Impact on Less-Educated Men: U.S.-born male workers with a high school degree or less in affected sectors saw a 1.3% reduction in their employment rate.
• Linked Job Loss: The data suggests a correlation where for every six lost male likely undocumented workers, one male U.S.-born worker also loses their job.

4. Economic Mechanism: Complementarity over Substitution

A central conclusion regarding the "why" of these results is that undocumented and U.S.-born workers are complements rather than substitutes.

• Production Interdependence: Because these workers often perform different but mutually dependent tasks, the loss of undocumented labor contracts overall labor demand in a sector.
• Sector-Level Correlation: The harm to U.S.-born workers was most pronounced in sectors like construction, which saw the largest drops in undocumented labor supply, further supporting the complementarity model.

Summary of the Broader Context

In the larger context of Trump 2.0, the authors conclude that the current enforcement regime is uniquely disruptive due to its scale and indiscriminate nature. They argue that the policy is undermining the economic prospects of the very workers it was intended to protect by shrinking the industrial sectors—such as agriculture and construction—that rely on a stable, multi-tiered workforce.