Choosing A Standard Population

Why Choose a Standard Population?

Rates in two groups can differ partly because the groups have different compositions. For example, one group may contain a larger share of older people, who may have higher event rates than younger people. A comparison that aims to separate rate differences from compositional differences needs a common set of stratum weights. Those weights come from a standard population: a specified target population whose distribution across strata supplies the reference composition.

Choosing the standard population is a substantive decision, not a calculation that can be postponed until after examining the study sample. The choice answers the question, “To the distribution of which population should the comparison refer?” The target might be a national population in a specified year, a regional population, or a policy-relevant population defined in advance. Once chosen, its composition—not the relative numbers recruited or observed in each study group—determines the weights.

Definition of Standard-Population Weights

Suppose the population is divided into $K$ mutually exclusive and collectively exhaustive strata, such as age bands. Let $N_k^*$ be the number of people in stratum $k$ of the chosen standard population, and let $N^*=\sum_{k=1}^K N_k^*$ be its total size. The asterisk marks quantities belonging to the standard population rather than the study sample.

Definition: The standard-population weight for stratum $k$ is the proportion of the chosen standard population in that stratum: $$ w_k=\frac{N_k^*}{N^*}, \qquad N^*=\sum_{j=1}^K N_j^*. $$ Thus $w_k\geq 0$ and $\sum_{k=1}^K w_k=1$. A standard population may be described by its stratum counts or directly by its stratum proportions; either representation gives the same weights after normalization.

The superscript star is important. Study data also have stratum counts, say $n_k$, and corresponding sample proportions $n_k/\sum_j n_j$. These proportions describe the study sample. They are not standard-population weights unless the study sample was deliberately constructed to represent the chosen target distribution and that is the intended standard. In many studies, recruitment, eligibility, follow-up, or group membership makes the observed stratum proportions different from the target proportions.

For rate adjustment, the standard population must be specified on the same stratum scheme used to obtain the stratum-specific rates. If the standard is divided into age bands 0–49, 50–69, and 70+, but study rates are available only for 0–64 and 65+, the weights cannot simply be paired with those rates. The categories must be made compatible, or the data must support an alternative common stratification. The weights alone do not solve a mismatch in stratum definitions.

Worked Example: Calculating Weights From Standard-Population Counts

A reference population has 48,000 people aged 0–44, 32,000 aged 45–64, and 20,000 aged 65 or older. Calculate the standard-population weights for these three age strata.

First add the stratum counts to obtain the total standard-population size:

N^*=48{,}000+32{,}000+20{,}000=100{,}000.

Divide each count by this same total:

w_1=\frac{48{,}000}{100{,}000}=0.48,\qquad w_2=\frac{32{,}000}{100{,}000}=0.32,\qquad w_3=\frac{20{,}000}{100{,}000}=0.20.

The weights sum to $0.48+0.32+0.20=1.00$, as required. The chosen standard population therefore assigns 48% of its weight to ages 0–44, 32% to ages 45–64, and 20% to ages 65 or older. These proportions are the relevant output of the choice; the fact that the original counts total 100,000 is not itself essential.

Weights Belong to the Target, Not to a Study Group

Suppose rates are being compared between two treatment groups. Each group may have a different age distribution, and both distributions may differ from the reference population. If each group is assigned its own observed age proportions, the two resulting summaries do not use a common composition. Conversely, using one set of weights taken from a prespecified standard gives both groups the same reference distribution.

The choice of reference distribution also defines the population to which an adjusted comparison pertains. A standard weighted toward younger people answers a different target-population question from one weighted toward older people when stratum-specific rates vary with age. There is therefore no universally correct standard independent of the purpose of the analysis. Good practice is to state the standard’s source, year or definition, strata, and counts or proportions.

Worked Example: Separating Study Composition From Target Weights

A study compares two hospitals using two age strata. The study includes 900 younger and 100 older patients at Hospital A, and 400 younger and 600 older patients at Hospital B. The intended standard population contains 7,000 younger and 3,000 older people. Find the standard weights and compare them with each hospital’s sample proportions.

The standard has $7{,}000+3{,}000=10{,}000$ people. Its weights are

w_{\text{young}}=\frac{7{,}000}{10{,}000}=0.70,\qquad w_{\text{older}}=\frac{3{,}000}{10{,}000}=0.30.

Hospital A has $900+100=1{,}000$ patients, so its observed proportions are $900/1{,}000=0.90$ younger and $100/1{,}000=0.10$ older. Hospital B has $400+600=1{,}000$ patients, so its proportions are $400/1{,}000=0.40$ younger and $600/1{,}000=0.60$ older.

The standard weights are therefore $(0.70,0.30)$, not Hospital A’s $(0.90,0.10)$ and not Hospital B’s $(0.40,0.60)$. The data from each hospital are used to estimate the relevant stratum-specific rates; the stated target population supplies the common weighting distribution. This distinction prevents a group’s sample size or age mix from silently changing the target of the comparison.

Two Useful Properties of the Weights

The basic weight definition has several consequences that help detect errors and maintain a meaningful target when the stratum scheme changes.

Result 1 (Normalization and target-share interpretation): If $N^*>0$, the weights defined by $w_k=N_k^*/N^*$ are nonnegative, sum to one, and equal the corresponding shares of the standard population. Multiplying every standard-population count by the same positive constant leaves all weights unchanged.

Proof. Each $N_k^*$ is a nonnegative count, and $N^*=\sum_k N_k^*>0$, so $w_k=N_k^*/N^*\geq 0$. Summing across strata gives

\sum_{k=1}^K w_k =\sum_{k=1}^K\frac{N_k^*}{N^*} =\frac{\sum_{k=1}^K N_k^*}{N^*} =\frac{N^*}{N^*}=1.

The fraction $N_k^*/N^*$ is exactly the share of all standard-population members in stratum $k$. If every count is multiplied by a positive constant $c$, the new weight is $cN_k^*/\sum_j cN_j^*=cN_k^*/(cN^*)=N_k^*/N^*$. Thus the weights depend on the distribution, not the scale of the counts. $\square$

This result explains why a published standard can be given as counts, percentages, or proportions. For example, counts of 7,000 and 3,000 and proportions of 0.70 and 0.30 represent the same standard. It also provides a direct arithmetic check: if purported weights do not sum to one (up to rounding), something is wrong or a category has been omitted.

Result 2 (Combining strata): If a set $A$ of standard-population strata is combined into one broader stratum, its weight is the sum of the original weights: $$ w_A=\frac{\sum_{k\in A}N_k^*}{N^*}=\sum_{k\in A}w_k. $$

Proof. The combined stratum contains $\sum_{k\in A}N_k^*$ members of the unchanged standard population, while the total population remains $N^*$. Its weight is therefore

w_A =\frac{\sum_{k\in A}N_k^*}{N^*} =\sum_{k\in A}\frac{N_k^*}{N^*} =\sum_{k\in A}w_k.

This is a statement about population shares. It does not imply that stratum-specific rates can always be combined by adding or simply averaging them: a rate for a merged stratum requires the appropriate underlying event and exposure information. The weight identity only says how much of the standard population lies in the merged category. $\square$

Worked Example: Combining Adjacent Age Bands

A standard population has counts of 12,000 people aged 0–19, 18,000 aged 20–39, 30,000 aged 40–64, and 40,000 aged 65 or older. Calculate the original weights, then find the weight for ages under 40 and the weight for ages 40 or older.

The total is $12{,}000+18{,}000+30{,}000+40{,}000=100{,}000$. The four weights are therefore $0.12$, $0.18$, $0.30$, and $0.40$, respectively. Checking, $0.12+0.18+0.30+0.40=1.00$.

The combined under-40 weight is the sum of the first two weights:

w_{\text{under 40}}=0.12+0.18=0.30 =\frac{12{,}000+18{,}000}{100{,}000}.

The combined 40-or-older weight is

w_{\text{40 or older}}=0.30+0.40=0.70 =\frac{30{,}000+40{,}000}{100{,}000}.

The two broader weights still sum to one. This calculation is valid for the target’s population shares. It does not by itself establish that the study has enough information to estimate rates for these broader groups; that depends on how the study observations and exposure are recorded.

Choosing the Target Carefully

A defensible choice of standard population follows the question being asked. A national standard may support comparison to a broad public-health reference; a local standard may make a local planning question more relevant; and a fixed policy population may be appropriate when evaluating consequences for a specified community. If comparisons will be reported across several studies or over time, using the same standard can aid comparability. Changing the standard changes the weighting distribution and therefore changes the precise target of the summary.

Before calculating weights, specify the target, its source and time reference, and the strata. Check that the categories cover the entire standard population without overlap, that each count refers to the same target definition, and that the resulting weights sum to one. Then verify that study data support the same strata for the groups being compared. If a standard assigns positive weight to a stratum for which a study group has no usable exposure or rate information, the desired adjusted comparison may not be estimable from that group’s data without additional assumptions or a revised target.

Common Pitfall

Wrong reasoning: “Hospital A has 90% younger patients, so the younger standard weight should be 0.90. Hospital B has 40% younger patients, so it should use 0.40. These are the most representative weights because they come from the data.”

Correction: Those numbers are the hospitals’ observed compositions. They may describe each hospital accurately, but they do not create one common target population. If the goal is comparison at a fixed reference composition, choose that population first and calculate its weights from its own stratum counts. Here the specified standard had 70% younger people, so both hospitals use a younger-stratum weight of 0.70. Use of the study proportions instead would answer a different question and would allow group composition to affect the weighting scheme.

Key takeaway: Standard-population weights are normalized stratum counts from a stated target population. They are fixed by that target, sum to one, and must be kept distinct from the study sample’s stratum proportions.

Check Your Understanding

Answer each question by identifying the target distribution and showing the relevant calculation.

A standard population has 25,000 people in stratum 1 and 75,000 in stratum 2. What are its two weights, and how can you check them?
A study group is 60% in the younger stratum, while the chosen standard is 35% younger. Which proportion supplies the younger standard weight, and why?
If all standard-population counts are multiplied by 10, what happens to the weights? Show this algebraically.
Three adjacent strata have standard weights 0.08, 0.17, and 0.25. What is the weight of their combined category?
What should be checked if a standard gives positive weight to a stratum with no usable rate information in one study group?

What You'll Learn