Impact of the Great Recession on U.S. Cohort Fertility

 The sources outline a multi-faceted theoretical framework to explain how economic downturns like the Great Recession (GR) impact fertility, noting that prevailing theories often provide conflicting or nuanced predictions.

Economic Theory of Fertility

The prevailing economic theory, notably advanced by Gary Becker (1981), treats fertility behavior similarly to the consumption of a durable good. This perspective suggests two competing effects:

• Income Effect: As a "normal good," the demand for children should rise with household income.
• Quality/Quantity Trade-off: High-income couples may choose to have fewer children while investing more heavily in each child. Because of these competing factors, Becker's theory does not provide a clear, singular prediction of how individual household fertility will respond to changing economic circumstances.

Aggregate Fertility and Postponement

While individual predictions are ambiguous, economic theory offers firmer expectations regarding aggregate fertility during crises. Under this view:

• Credit Access: During recessions, credit costs rise and access declines, making it difficult to "borrow against future earnings" to fund the high cost of children.
• Pro-cyclical Response: Just as households delay purchasing homes or cars during a downturn, they are expected to postpone births until an economic recovery occurs.
• Transitory Nature: This reasoning suggests the primary effect of a recession is delay, making the response pro-cyclical and transitory, with only "second-order effects" on total lifetime fertility.

Fertility Preferences and Cohort Differences

Demographic and sociological theories place fertility preferences at the center. These theories posit that individuals have a "desired fertility" target. This lead to specific predictions regarding age:

• Target Achievement: There should be little to no recession effect for individuals who have already achieved or exceeded their desired number of children.
• Cohort Impact: Because younger women are further from their fertility targets than older women, recession effects are predicted to be greatest in the youngest cohorts and decrease in magnitude for older cohorts.

Proximate Determinants Framework

The proximate determinants of fertility framework (Davis and Blake, 1956) suggests that the causal effects of a recession do not act directly but operate through specific mechanisms. These "fertility proximates" include:

• Decreases in sexual activity.
• Improved or increased use of contraception and abortion.
• Changes in union formation, such as delays in marriage or potential increases in cohabitation as couples seek to share costs during hard times.

Formal Demographic Theory and Measurement

The sources also draw on formal demographic theory to justify their focus on cohort measures rather than period measures (like the Total Fertility Rate or TFR).

• Tempo Effects: Period measures can be distorted by "tempo effects"—short-term fluctuations caused by the timing (postponement) of births—which may not reflect actual changes in the number of children women have over their lifetimes.
• Stationarity: Differences between period and cohort measures emerge primarily in non-stationary contexts where the age structure of fertility is changing, such as the current U.S. trend where fertility is declining for women in their 20s but increasing for those in their 30s.

The sources introduce a novel cohort discontinuity design specifically developed to obtain credibly causal estimates of the Great Recession’s (GR) impact on fertility. This design moves beyond simple associations by identifying effects for single-year birth cohorts, as well as by race.

Core Identification Strategy

Causal identification in this design rests on two critical, untestable assumptions:

• Exogeneity of Timing: The assumption that the start (December 2007) and end (May 2009) of the GR could not be anticipated by households. The sources note that the official NBER announcement of the recession's end came 17 months after it actually occurred, supporting the idea that individuals could not have planned their fertility around its specific timing.
• Projection Assumption: The assumption that demographic methods can accurately project what a cohort's fertility would have been (the counterfactual) had the recession not occurred. This is considered more credible for short-term projections (like the 18-month duration of the GR) than for long-term lifetime fertility.

Comparison with Other Causal Designs

The authors distinguish their approach from three common causal inference frameworks:

• Difference-in-Differences (DiD): DiD was deemed not feasible because it requires a control group that remains untreated. In the case of the GR, the authors presume that all women were "treated" by the economic shock simultaneously.
• Regression Discontinuity (RD): While similar in using "local linear regressions" near a threshold, RD typically uses a "running variable" (like a test score) to allocate units into treatment and control groups. The cohort discontinuity design instead exploits demographic regularity, assuming that fertility trends behave smoothly across adjacent cohorts.
• Interrupted Time Series (ITS): Unlike ITS, which uses a unit’s own past data to project its future, this design uses the observed fertility of older cohorts to project the counterfactual for younger ones.

Estimation and Counterfactuals

To estimate the effect for a specific cohort (e.g., women born in 1989), the design uses data from five older, adjacent cohorts (those born in 1984–1988).

• Local Linear Regressions: These are used to project the counterfactual fertility rate for the "treatment" cohort.
• Monthly Fixed Effects: The model includes 18 monthly fixed effects to account for birth seasonality and the overlapping age structure of birth cohorts.

Placebo Analysis and Bias Adjustment

Because local regressions can be "noisy," the authors implemented a placebo-adjusted estimation procedure. They ran 39 "pseudo-treatments" using 18-month periods prior to the actual recession. By calculating the median bias from these placebo tests, they could adjust their final estimates to account for any systematic departures from linear trends.

Statistical Inference

The design uses permutation methods (proposed by R.A. Fisher) rather than traditional frequentist methods to construct confidence intervals. This is because the study analyzes all U.S. births (a population) rather than a sample, making the frequentist notion of "repeated sampling" inappropriate. Permutation tests are exact and distribution-free, relying on the concept of "exchangeability" to determine if the observed decline is statistically significant.

Design Limitations

The sources acknowledge that while the design is robust for identifying short-run causal effects (the "effect of the cause"), it has limitations:

• Long-Run Effects: It is not well-suited for determining if the fertility decline was permanent or merely a postponement, as long-term projections are less credible.
• Causal Pathways: The design identifies the total effect of the recession but cannot distinguish between specific mechanisms, such as whether the decline was driven by high unemployment versus general economic uncertainty.

The empirical findings from the sources provide strong evidence that the Great Recession (GR) caused a pro-cyclical decline in fertility across nearly all single-year cohorts of U.S. women born between 1964 and 1992. These findings highlight a clear age gradient and significant racial heterogeneity in how the economic shock affected childbearing.

Overall Impact and Age Gradient

The study’s preferred placebo-adjusted estimates reveal that the GR’s impact was most severe for the youngest women.

• Youngest Cohorts: Women in their teens and 20s during the recession experienced the sharpest declines. For the 1992 birth cohort (aged 17 in 2009), the relative decline was as high as 30.5%.
• Older Cohorts: The magnitude of the effect generally decreased with age. For cohorts born between 1964 and 1976 (aged 33–45 in 2009), relative effects fluctuated between +0.1% and -5.1% and were largely not statistically significant.
• Statistical Significance: The estimated declines were statistically significant at the .01 level for women born between 1965 and 1992 (aged 17–44 in 2009).

Absolute vs. Relative Magnitude

The sources emphasize a distinction between relative percentage changes and absolute changes in births per woman.

• Relative Declines: For women in their teens and 20s, the relative declines ranged from 7% to 30%.
• Absolute Declines: These large relative percentages represent more modest absolute changes, ranging from 0.013 to 0.031 births per woman.
• Context: This discrepancy occurs because the absolute changes were measured against low baseline fertility levels in younger cohorts. The largest absolute decline occurred in the 1988 cohort, with a reduction of 0.0305 births per woman.

Racial Heterogeneity

The recession's impact varied considerably by race, with Black women experiencing significantly larger declines than White women.

• Black Women: Relative declines reached -35.1% for the 1992 cohort, with effects remaining statistically significant for all cohorts born between 1976 and 1992. The largest absolute decline for Black women was 0.0435 births per woman (1990 cohort).
• White Women: Relative declines for the 1992 cohort were -28.5%, and effects were statistically significant for those born between 1980 and 1992. The largest absolute decline for White women was 0.0218 births per woman (1988 cohort).

Long-Term Cumulative Fertility

A final set of analyses followed the most affected cohorts forward to assess whether the GR caused permanent reductions or temporary postponements.

• Observed Fertility (2022): By 2022, cumulative births per woman for these cohorts ranged from 1.15 (1992 cohort) to 2.14 (1980 cohort).
• Projected Fertility (2037): Projections to 2037 suggest that even for the hardest-hit cohorts, lifetime fertility is likely to remain at or near replacement levels. Projected levels range from 1.89 for the 1992 cohort to 2.16 for the 1981–1983 cohorts.
• Conclusion on Duration: While the evidence is "suggestive" rather than causal for the long run, it implies the 18-month recession was a relatively short period compared to a woman's total reproductive years, allowing for potential recovery.

Robustness and Sensitivity

Sensitivity analyses confirmed the stability of these findings across different modeling assumptions. The results remained robust when the researchers:

• Varied the start and end dates of the GR by two months.
• Used different numbers of pretreatment cohorts (four or six instead of five).
• Applied different weighting kernels (triangular or Gaussian).
• Used reported gestational age instead of a fixed nine-month assumption.

The sources provide a descriptive, "suggestive" analysis of cumulative fertility to determine if the Great Recession (GR) caused permanent reductions in lifetime childbearing or merely temporary postponements. By following the cohorts most affected by the recession—women in their teens and 20s during the downturn—the authors assess both observed progress and future projections.

Observed and Projected Cumulative Fertility

The researchers tracked the most-impacted cohorts (born 1980–1992) through the year 2022 and then projected their lifetime fertility to 2037 using established demographic methods.

• Observed Fertility (2022): By 2022, cumulative births per woman for these cohorts varied significantly based on age. The 1992 cohort (aged 30 in 2022) had an average of 1.15 births, while the 1980 cohort (aged 42 in 2022) had reached 2.14 births.
• Projected Lifetime Fertility (2037): Projections suggest that despite the recession's impact, these women are likely to end their reproductive years with fertility levels at or near replacement (traditionally set at 2.10 births per woman). Specifically:
  • The 1992 cohort is projected to reach 1.89 births per woman.
  • The 1981–1983 cohorts are projected to reach 2.16 births per woman.
• Potential Underestimation: The authors note that the 1.89 projection for the 1992 cohort may be conservative. Their projection model "freezes" fertility rates after a certain point, but current U.S. trends show increasing fertility for women in their late 30s and 40s, which could push these final numbers higher.

Interpretation: Postponement vs. Permanent Reduction

The findings suggest that the 18-month duration of the Great Recession was relatively short when measured against a woman's total reproductive life span. This supports the idea that the primary effect of the recession was to delay or postpone births rather than to prevent them entirely. Even for cohorts that experienced relative fertility declines as high as 30% during the recession, the absolute reduction (roughly 0.013 to 0.031 births per woman) was small enough that it did not fundamentally alter their ability to reach near-replacement fertility levels over the long run.

Causal Caveats and Limitations

The authors are careful to distinguish these cumulative results from their causal findings regarding the 18-month recession period itself.

• Non-Causal Evidence: While the design provides credibly causal estimates for the recession's immediate impact, the cumulative analysis is described as purely descriptive and only "suggestive".
• Credibility of Projections: Causal identification of long-term effects is difficult because long-range projections are far less credible than the short-term projections used to identify the initial recession shock.
• Counterfactual Challenges: A truly causal assessment of lifetime fertility would require a counterfactual of what a cohort's completed fertility would have been across several decades had the GR never occurred, which the authors view as a major hurdle for current demographic methods.

The sources identify several key limitations of the study’s novel cohort discontinuity design, emphasizing that while it provides strong evidence for short-term causal effects, it faces challenges regarding long-term projections, specific causal mechanisms, and data constraints.

Short-Run vs. Long-Run Effects

The most significant limitation of the design is that it is primarily suited for identifying short-run effects during the specific 18-month duration of the Great Recession.

• Postponement vs. Permanence: Because the design relies on projecting counterfactuals based on the fertility of adjacent older cohorts, it cannot definitively determine whether the observed fertility declines were permanent (leading to fewer lifetime births) or transitory (postponements followed by recovery).
• Projection Credibility: Identifying long-term effects would require projections extending years into the future, which the authors view as "far less credible" than the short-term projections used for the initial 18-month shock.

Untestable Identification Assumptions

Causal identification within this framework rests on two critical but untestable assumptions:

1. Exogeneity of Timing: The assumption that the specific start and end dates of the recession could not be anticipated by households.
2. Projection Assumption: The assumption that demographic methods can accurately project what a cohort's fertility would have been in a counterfactual scenario where the recession never occurred.

Inability to Isolate Causal Mechanisms

The study identifies the "effects of causes" rather than the "causes of effects".

• Confounded Factors: The design identifies the total impact of the recession period but cannot distinguish the effect of the recession itself from correlated factors such as high unemployment or heightened economic uncertainty.
• Proximate Pathways: Similarly, while the authors acknowledge that fertility is governed by proximate determinants (such as changes in sexual activity, contraception use, or union formation), their design is not equipped to identify which specific causal pathway drove the decline.

Data and Scope Limitations

The authors also note several limitations stemming from the data and the standard demographic focus of the study:

• Parity: Due to data constraints, the analyses refer to the effects on births at all parities. This means the study cannot estimate whether the recession affected first births differently than second or subsequent births.
• One-Sex Process: Following standard demographic practice, the study treats fertility as a "one-sex process," focusing exclusively on women. This limitation risks overlooking how recessions affect males and their specific fertility-related behaviors.
• Denominator Construction: There are also inherent data limitations related to the construction of the population counts (denominators) needed to calculate monthly cohort fertility rates.