Tempo Effects in Mortality

Definition

Tempo effects refer to distortions in demographic period measures caused by shifts in the timing of events (births, deaths). In fertility, the total fertility rate (TFR) is distorted downward when birth timing shifts later (women defer births, reducing the count in any single year). The debate over whether a similar tempo effect distorts period life expectancy $e_0$ was central to a 2002–2005 methodological controversy. The resolution: $e_0$ is not subject to tempo distortion because it is already a timing measure, not a count measure.

Key Ideas

Quantum vs. tempo. Quantum measures the intensity or count of a demographic event (e.g., average number of births per woman). Tempo measures the timing (e.g., mean age at childbirth, or mean age at death). Period measures of quantum — like TFR — are distorted by tempo shifts because an acceleration in births bunches them in a given year, artificially inflating the period count, and a deceleration depresses it.
TFR tempo distortion (Bongaarts and Feeney 1998). TFR is a quantum measure: it counts births per woman-year. If women shift births to earlier ages (a tempo shift), TFR rises mechanically in that period even if lifetime fertility (completed fertility) is unchanged. The Bongaarts-Feeney formula adjusts TFR by removing this timing artifact, yielding a "tempo-adjusted TFR" closer to completed fertility.
$e_0$ is a tempo measure, not a quantum measure. Period life expectancy measures the mean age at death of the synthetic cohort — it is directly a measure of timing. There is no count of deaths that could be artificially inflated or deflated by timing shifts. When mortality improves (deaths shift to later ages), $e_0$ rises to reflect that shift — this is precisely what it is supposed to do, not a distortion. Wilmoth (2005) formalizes this: tempo-adjusting $e_0$ converts it from a synthetic-cohort measure to the Cross-Sectional Average Length of Life (CAL), which is a valid transformation but not a correction of a distortion.
The Bongaarts-Feeney claim and its limits. Bongaarts and Feeney (2002, 2003) argued that period $e_0$ overstates the "true" life expectancy because improving mortality artificially shifts deaths out of the current period, inflating $e_0$ . Wilmoth (2005) and Guillot (2003, 2005) show this argument is incorrect: tempo-adjusted $e_0$ under their generalized formula equals CAL, a population-dynamics measure. CAL is lower than period $e_0$ under improvement, but this reflects the fact that CAL and $e_0$ measure different things, not that $e_0$ is biased.
No distortion ≠ no gap. Accepting that $e_0$ is not tempo-distorted does not imply that $e_0$ and cohort life expectancy (LE) are equal. Under improving mortality, period $e_0$ systematically exceeds cohort LE (because period $e_0$ uses current-favorable rates while cohort LE uses past-higher rates). This gap is real and interpretable — it just is not a bias. See Period vs. Cohort Life Expectancy.
Average Cohort Life Expectancy (ACLE) as a tempo-robust alternative to CAL (Schoen and Canudas-Romo 2005): If CAL fails as a tempo correction (it does — see Cross-Sectional Average Length of Life (CAL)), what measure is actually insensitive to timing? Schoen and Canudas-Romo propose ACLE: a survivorship-weighted mean of all active cohorts' completed life expectancies (LEs). Because ACLE depends only on completed cohort LEs — not on when within each cohort deaths occur — pure timing shifts that leave cohort LE unchanged leave ACLE exactly flat. In simulations with timing oscillations, ACLE is the smoothest of the four aggregate measures (cohort LE, period LE, CAL, ACLE). See Average Cohort Life Expectancy (ACLE).
CAL is the result of the "adjustment." Applying the Bongaarts-Feeney generalized correction to $e_0$ yields CAL. CAL is interpretable and valuable as a population-dynamics measure in its own right, but it is not a "true" or "debiased" life expectancy — it is a different measure with a different question. See Cross-Sectional Average Length of Life (CAL).

How It Works

The tempo distortion in TFR arises mechanically: $\text{TFR} = \int f(a,t)\, da$ where $f(a,t)$ is the age-specific fertility rate. If births shift forward by $d$ years at all ages simultaneously, $\text{TFR}(t) \approx \text{completed fertility} / (1 - \dot{r}_\text{age})$ where $\dot{r}_\text{age}$ is the rate of change in mean age at birth. This adjustment is well-defined because TFR is a period integral of rates, not a timing quantity.

For mortality, the analogous quantity would be age-standardized death rates or crude death rates — which are indeed affected by age-structure changes and timing shifts. But $e_0$ is not a rate or a count — it is already the mean age at death for the synthetic cohort. Under declining mortality, period $e_0$ rises because mean age at death for the synthetic cohort rises. There is no artifact to remove.

Why It Matters

Methodological clarity. The tempo-bias debate in mortality was prominent enough to generate a special symposium in Demographic Research (2004–2005). Resolving it correctly matters for interpretation: forecasts of future $e_0$ , assessments of longevity gains, and comparisons across countries should use period $e_0$ as a valid period measure, not treat it as a distorted proxy for something else.
Policy relevance. If period $e_0$ were genuinely biased upward, policy conclusions drawn from it (e.g., "longevity has improved substantially") would be overstated. Wilmoth's result confirms that measured improvements in $e_0$ are real, not artifacts of timing shifts.

Open Questions

The distinction between tempo effects in fertility (where the effect is genuine) and mortality (where Wilmoth argues it is not) rests on the count vs. timing nature of TFR vs. $e_0$ . Whether other mortality summary measures (e.g., standardized death rates, years of life lost) are subject to analogous tempo effects is not fully resolved.