Researchers have identified 3-parameter Generalized Gamma model as the most accurate statistical tool for analysing first birth intervals in North-East India, where the median interval of just 19 months falls below the WHO’s 24-month safe threshold, raising urgent maternal and child health concerns.
A peer-reviewed study published on December 1, 2025 in the National Journal of Community Medicine (Vol. 16, Issue 12) has pinpointed a key statistical weakness in how birth spacing is currently analysed in one of India’s most demographically complex regions — and offers a measurably superior solution. The paper, titled “Parametric Models for First Birth Interval in North-East India: Identifying a Suitable Model,” was authored by Chanambam Rajiv Mangang and Kshetrimayum Anand Singh of the Department of Statistics, Manipur University.
The study concludes that the three-parameter Generalized Gamma model is the most accurate parametric tool for modelling the “First Birth Interval” (FBI) — defined as the duration between a woman’s marriage and her first live birth — across the eight states of North-East India. The finding carries direct implications for public health planning in a region where adolescent marriages are common and ethnic diversity is high.
Drawing on data from the National Family Health Survey (NFHS-5, 2019–20), the researchers analysed a large-scale sample of 60,820 ever-married women aged 15–49 across all eight North-Eastern states. The headline finding is stark: the median first birth interval in North-East India is just 19 months — significantly shorter than the national Indian average of 23 months.
This figure places much of the region’s reproductive-age population in a high-risk category. Under World Health Organization (WHO) guidelines, any birth interval shorter than 24 months is associated with elevated risks of maternal and infant mortality. Short intervals can trigger “maternal depletion,” a condition in which a mother’s body does not fully recover between pregnancies, adversely affecting her health and the survival prospects of her newborn.
The researchers also noted a compounding demographic effect: because the FBI is a primary determinant of a woman’s completed fertility — that is, the total number of children she will have — shorter intervals contribute directly to higher fertility rates and intensified population growth pressures across the region.
Standard epidemiological practice often relies on the Cox proportional hazards model, a non-parametric approach. However, the authors argued that fully parametric models offer decisive advantages for public health planning: they produce more precise forward-looking predictions and clinically meaningful estimates of survival time, making them better suited for designing and evaluating interventions.
The research team systematically compared five competing parametric models:
• Weibull — a widely used baseline model
• Log-normal — suited to skewed time-to-event data
• Gompertz — commonly applied in demographic survival analysis
• Gamma — a flexible two-parameter distribution
• Generalized Gamma — a three-parameter distribution encompassing all of the above
To determine the best-fitting model, the team applied the Akaike Information Criterion (AIC) — a standard statistical measure where a lower value signals a better balance of accuracy and model simplicity — alongside graphical diagnostics including density curve overlays and Q-Q (quantile-quantile) plots.
The Generalized Gamma distribution emerged as the clear winner across all three evaluation criteria:
1. Lowest AIC Score
The Generalized Gamma model recorded an AIC value of 423,085.4 — the lowest among all five models tested. A lower AIC indicates the model captures the data’s complexity without overfitting, offering the most reliable basis for future predictions.
2. Best Visual Alignment with Observed Data
When density curves for each of the five models were superimposed over histograms of real-world birth interval data, only the curve representing the Generalized Gamma distribution visually matched the observed patterns of first birth intervals across the North-East. The other models exhibited notable deviations from the actual distribution.
3. Superior Q-Q Plot Performance
Through Q-Q (quantile-quantile) plot diagnostics, the researchers found that the Generalized Gamma model followed the reference line “perfectly up to a reasonable level,” confirming that it accurately captured the shape, scale, and skewness of the real-world data. The authors also addressed a common statistical challenge: formal goodness-of-fit tests such as the Kolmogorov-Smirnov test showed significant deviations, but they noted that such tests become “overly sensitive” in very large datasets and are less reliable than graphical diagnostics in this context.
Notably, the Gompertz model produced a median estimate of 19.46 months — the closest numerically to the observed 19-month median — yet it failed to match the overall shape of the data as effectively as the Generalized Gamma, underscoring that a single summary statistic is insufficient for evaluating model fit.
The defining strength of the Generalized Gamma model lies in its structural flexibility. Governed by three parameters — location, scale, and shape — the distribution can mathematically mimic the Weibull, Log-normal, and standard Gamma distributions as special cases. This means it can represent
• increasing risk — where birth probability rises over time after marriage;
• decreasing risk — where it falls; or
• hump-shaped (bell-curve) risk — where birth probability peaks at a certain interval and then declines.
This adaptability is critical for modelling human reproductive events. Unlike disease onset or mechanical failure, birth intervals do not follow a simple linear hazard pattern; the probability of a first birth changes in complex, non-monotonic ways across the months following marriage. No simpler two-parameter model can capture this nuance as effectively.
The researchers were explicit that their work is not a purely academic exercise. The accurate modelling of FBI patterns provides health administrators with a robust predictive tool to
• identify sub-populations at highest risk of short birth intervals;
• simulate the likely impact of proposed interventions before deployment; and
• target community-level programmes — such as premarital counselling and family planning education — at couples most likely to experience intervals below the 24-month WHO threshold.
As the authors concluded: “The Generalized Gamma model can inform policies to extend first birth intervals, reducing risks of adverse maternal and child health outcomes.”
The study’s current iteration used null models — that is, models without covariates such as educational attainment, economic status, or ethnicity. The authors acknowledged this as a limitation and indicated it as a direction for future research, noting that incorporating these predictors into the Generalized Gamma framework would yield more granular and actionable public health recommendations for the North-East’s diverse populations.
In a region where adolescent marriages remain prevalent and access to reproductive health services is uneven, a statistically sound model for predicting first birth intervals is not merely a methodological preference — it is a public health necessity. This research from Manipur University provides precisely that foundation.
(Keithellakpam Manikanta Meetei is a seasoned journalist and a former educator. He also writes under his pen name Keicha Chingthou Mangang instead of his actual name. You can contact him at chingthouheiya@gmail.com)
