技术需求：填补美国普查数据中缺失的分年龄离散人口值

Got it, let's walk through how to fill in those missing discrete age-group population numbers from your Census dataset. This is a common challenge with historical demographic data, and there are a few reliable approaches to choose from depending on your accuracy needs and access to supplementary data.

Method 1: Age-Specific Mortality Model (Most Accurate) #

This is the gold standard if you can get your hands on corresponding year life tables—Census.gov and the CDC's NCHS have these readily available for the US. The idea is to split the aggregated high-age total using survival probabilities from the life table.

Steps: #

• Grab the life table for the target year (e.g., 1940's life table for 85+ age groups, 1980's for 100+). Extract survival ratios or remaining life expectancy for each age above your cutoff.
• Calculate weights for each age group: divide the survival ratio of that age by the sum of survival ratios for all ages in the aggregated group.
• Multiply the aggregated total (e.g., 85+ sum for 1940) by each weight to get the imputed discrete population for that age.

Example Code (Python): #

import pandas as pd

def impute_via_life_table(total_aggregated, life_table_df, cutoff_age):
    # Filter life table to only ages we need to impute
    high_ages = life_table_df[life_table_df["age"] >= cutoff_age].copy()
    # Calculate weight based on survival ratios (adjust if using death probabilities)
    high_ages["weight"] = high_ages["survival_ratio"] / high_ages["survival_ratio"].sum()
    # Impute population
    high_ages["imputed_pop"] = high_ages["weight"] * total_aggregated
    # Round to nearest integer (population counts are whole numbers)
    high_ages["imputed_pop"] = high_ages["imputed_pop"].round().astype(int)
    # Final check: adjust last age if sum doesn't match aggregated total
    sum_diff = total_aggregated - high_ages["imputed_pop"].sum()
    high_ages.loc[high_ages["age"] == high_ages["age"].max(), "imputed_pop"] += sum_diff
    return high_ages[["age", "imputed_pop"]]

# Usage example for 1940's 85+ group
# 1940_life_table = pd.read_csv("1940_us_life_table.csv")
# 1940_85_plus_imputed = impute_via_life_table(1940_total_85_plus, 1940_life_table, 85)

Method 2: Exponential Decay (Quick & No Extra Data Needed) #

If you don't have access to life tables, this is a simple fallback. High-age populations follow a roughly exponential downward trend, so you can extend the pattern from the last available discrete age groups.

Steps: #

• Calculate the average decay rate using the last 3-5 available high-age groups (e.g., for 1940, use ages 80-84). The decay rate is pop_current / pop_previous for each consecutive age.
• Start from the cutoff age (85 for 1940) and multiply the previous age's population by the average decay rate. Repeat until you've covered a reasonable upper age limit (e.g., 110, since counts get negligible beyond that).
• Adjust the final age group's count to ensure the sum of all imputed ages matches the original aggregated total.

Method 3: Cross-Year Interpolation (For Intermediate Years) #

If you have full discrete data for adjacent years (e.g., 1980 and 2000 have 100+ discrete counts), you can interpolate to fill in gaps for years like 1990.

Steps: #

• For each high age (100, 101, etc.), pull the population counts from the years before and after your target year.
• Use linear interpolation (or time-series smoothing if you have more data points) to estimate the count for your target year.
• Scale all interpolated values slightly if their sum doesn't match the target year's aggregated total.

Critical Validation Step #

No matter which method you use, always:

1. Check that the sum of your imputed discrete values exactly matches the original aggregated total for the high-age group.
2. Spot-check trends across years (e.g., 1950's 85-year-old population should be larger than 1940's, reflecting overall population growth) to ensure your imputations make demographic sense.

内容的提问来源于stack exchange，提问作者Threadid