Descriptive Regression Questions and Conditional Means

Describing Average Outcomes Across Predictor Values

A descriptive regression question asks how average outcomes vary across values of a predictor in a defined population. For example, a health service might ask: among patients attending its clinics this year, what is the average waiting time for visits at clinics in each region? The aim is to describe the population’s pattern, not to claim that region causes a particular waiting time.

The target must be clear about three things: the population being described, the outcome being averaged, and the predictor values or groups being compared. A statement such as “How does waiting time vary by region among visits to these clinics in 2026?” is more precise than “Does region matter?” The first specifies a descriptive target; the second is ambiguous about both the target and what “matter” means.

A conditional mean describes an average, not the outcome of every individual in that group. If the mean waiting time in a region is 20 minutes, individual visits may have shorter or longer waits. Nor does the conditional mean alone describe how spread out the outcomes are.

For a categorical predictor, the meaning is direct: identify the members of each category in the target population and average their outcomes. For a quantitative predictor that takes countable values, such as number of visits, the same idea applies at each value with population members. For a continuous predictor, an exact value may occur with probability zero, so “individuals with exactly \(X=x\)” may not be a literal subgroup. In that setting \(E(Y\mid X=x)\) is understood as the mean specified by the conditional distribution of \(Y\) given \(X=x\), where that conditional distribution is defined. It describes the mean response at that predictor value; it is not obtained by dividing by the probability of the event \(X=x\).

The target is defined only where the population has relevant predictor values and the conditional mean exists. If no one in the target population belongs to a category, its group mean is not defined. Likewise, a mean at a predictor value outside the population’s range cannot be described from that population’s conditional means without making additional assumptions.

Population Targets and Sample Estimates

The conditional mean is a population quantity. In practice, a study often observes a sample from the target population and calculates a sample mean within each observed group or predictor value. That calculated mean is an estimate of the corresponding population conditional mean, not the population quantity itself.

A descriptive question does not automatically require a regression model. One can define the population target and summarize sample means without assuming a straight-line relationship or any other particular shape. If a model is later used to summarize or estimate a pattern, its assumptions need separate consideration. For the descriptive target itself, the essential choices are the population, outcome, predictor, and relevant predictor values.

The scale and units matter. If \(Y\) is waiting time measured in minutes, each conditional mean is in minutes. A comparison between two such means is also in minutes. If the outcome is a score measured in points, the means and differences are in points. Stating the units helps make the target interpretable and guards against describing a change in the predictor as though it were a change in the outcome.

Patterns Across Predictor Values

A descriptive question can concern a quantitative predictor as well as categories. The key is to state which values are being described and to interpret the conditional means as average outcomes at those values. A pattern that increases, decreases, or varies irregularly can be a useful description, but the pattern alone does not explain why outcomes differ.

How Group Means Combine

The overall mean of an outcome is generally not the simple average of the group means. Groups may have different numbers of observations, so each group contributes in proportion to its size. The same arithmetic identity applies to population means and their group-specific conditional means, using population group sizes or proportions as weights.

To see why, suppose a sample is divided into groups. In group \(g\), let \(n_g\) be the number of observations and let \(\bar{y}_g\) be its mean. The sum of outcomes in that group is \(n_g\bar{y}_g\), since rearranging the mean formula gives \(\sum_{i:G_i=g}y_i=n_g\bar{y}_g\). Add these group sums to obtain the total outcome sum, then divide by the total sample size.

Thus, the overall mean is a weighted average of group means, with group sizes as weights. A group with more observations contributes more to the pooled mean. The overall mean answers a different question from the list of conditional means: it averages across the predictor distribution in the sample or population rather than describing the mean at each predictor value.

Interpreting the Target Without Overclaiming

A descriptive regression question is useful when the goal is to understand how outcomes are distributed across a predictor in a defined population. Conditional means can reveal differences between categories or changes in average outcome across quantitative predictor values. They also make the question concrete: rather than asking vaguely whether a predictor “matters,” one can name the average outcome of interest at each relevant value.

The population specification is important. “Mean waiting time by region” could refer to all visits in one year, visits at a particular service, or visits in a different region or period. Those are different target populations and may have different conditional means. The sample’s study units and selection process must also align with the population the analyst intends to describe. A sample mean is not automatically representative merely because it can be calculated.

Keep three statements distinct: the target is a population conditional mean; the estimate is a sample conditional mean; and a causal claim says something about what would happen under an intervention. A difference between sample means directly describes the observed sample. It estimates a population difference only when the sampling and measurement support that generalization. Neither statement by itself says what would happen if the predictor were changed.

Common Mistakes

• Treating the sample mean as the population target. The sample mean is calculated from observed data; it estimates the population conditional mean. Keep those quantities distinct.
• Leaving the population unspecified. “Average score by study hours” is incomplete without stating whose scores and which time period the target concerns.
• Confusing a conditional mean with an individual outcome. A group mean is an average; it does not predict that every member has that value.
• Calling a descriptive difference causal. Different conditional means describe how averages vary across groups or values. They do not, by themselves, show that changing the predictor would change the outcome.
• Averaging group means without considering group sizes. The pooled mean weights each group by its number of observations or, for a population quantity, its population share.
• Describing values unsupported by the target population. A conditional mean is not defined for an empty category, and a value outside the observed population range requires additional assumptions to describe.