Why the Question’s Purpose Matters
Regression can be used in different ways. A researcher might summarize a pattern in recorded data, investigate whether two variables are related, or use available information to forecast an outcome for a case. These tasks can involve similar measurements, but they do not ask the same question and do not justify the same conclusion.
Before choosing an analysis, identify the purpose of the research question. Is it asking what is observed, how or why something is related to something else, or what outcome to expect? These correspond to descriptive, explanatory, and predictive questions. The distinction is about the intended conclusion, not simply the topic or the statistical method.
Three Purposes, Three Kinds of Conclusion
A descriptive conclusion reports what the data show. It might state that students who spent more time studying tended to earn higher quiz scores in a particular set of records. The conclusion is about the observed pattern as described; it does not, by itself, say what produced the pattern.
An explanatory conclusion concerns a relationship: whether differences in one variable are associated with differences in another, or whether a proposed factor helps account for an outcome. The word “explanatory” names the purpose of the question, not a guarantee that the analysis has established a cause. A relationship found in data may be useful for understanding, but a causal claim requires evidence and reasoning that support causation. Association alone is not enough.
A predictive conclusion concerns expected outcomes for cases. For example, information about a student’s prior quiz results might be used to forecast that student’s next score. The focus is on how well the prediction performs for cases to which it is applied. A successful forecast does not necessarily identify why the outcome occurred.
| Question type | Main aim | Typical conclusion |
|---|---|---|
| Descriptive | Summarize what is observed | A report of values or patterns in the data |
| Explanatory | Understand a relationship | A conclusion about how variables are related, with causal wording only when warranted |
| Predictive | Forecast an outcome for a case | An expected value or forecast, whose performance can be assessed |
These purposes are not defined by whether a question contains numbers, uses the word “relationship,” or is analyzed with regression. Instead, read for the job the answer is meant to do. A question asking for the average score in a recorded group is descriptive. Asking whether study time is related to score is explanatory. Asking what score to expect for a particular student is predictive.
Classifying a Question Step by Step
Look for whether the question asks for a summary, an account of a relationship, or an expected outcome for a case.
A summary points to description; understanding a relationship points to explanation; forecasting points to prediction.
Do not turn a descriptive pattern into a cause, or treat a forecast as an explanation of why a case has its outcome.
Worked Example: Three Questions About Study Time and Scores
Suppose, purely hypothetically, that a teacher has records for 80 students, including the number of hours each student studied before a quiz and the score each student earned. Classify each question and state the kind of conclusion it supports.
Question 1: “What scores did these 80 students earn, and how were the scores distributed?” This is descriptive. The answer summarizes the recorded scores, perhaps with a typical score and a description of how scores vary. It says what appeared in these records; it does not explain why students received those scores.
Question 2: “Is study time related to quiz score among these students?” This is explanatory. It asks about a relationship between the two recorded variables. A supported conclusion could describe whether greater study time tends to accompany higher or lower scores. The records alone do not show that studying more caused any score difference; other factors could also be related to both.
Question 3: “Given a student’s recorded study time, what quiz score should the teacher expect for that student?” This is predictive. The answer is a forecast for a case based on available information. Its usefulness depends on how well such forecasts perform for the students or situations in which they will be used. It does not, by itself, explain why a student earns a particular score.
Decision: The setting and measurements are the same in all three questions, but their purposes differ. The first requests a summary, the second asks about a relationship, and the third requests a forecast. Therefore, the conclusion must be framed to match the question, not merely the data available.
Explanatory Questions and Causal Language
In ordinary language, “explain” can mean “give a cause.” In statistical work, however, questions called explanatory may ask about association without establishing causation. A question such as “Are students who attend more review sessions likely to earn higher scores?” can be studied as a relationship. It does not automatically support the stronger statement “Attending more sessions raises scores.”
The distinction matters because a variable can help account for or predict differences without being the cause of those differences. For instance, students who attend many review sessions might also differ in prior preparation, available time, or motivation. If so, a relationship between attendance and scores cannot by itself separate the contribution of attendance from those other differences.
When reading an explanatory question, check the wording of the conclusion it invites. If the question asks whether variables are related, report an association. If it asks whether changing one factor would change an outcome, that is a causal claim and needs suitable evidence. Do not let the word “explanatory” silently turn an observed association into a causal result.
Worked Example: A Relationship Is Not Automatically a Cause
Imagine a hypothetical city dataset showing that neighborhoods with more public transit stops tend to have fewer cars per household. A planner asks, “How is the number of transit stops related to car ownership?”
This is an explanatory question because it asks about a relationship between two variables. If the records show that neighborhoods with more stops tend to have fewer cars per household, the conclusion can describe that association in those data.
Now suppose someone changes the question to, “Would adding transit stops reduce car ownership?” This asks for a causal conclusion: it concerns what would happen if transit availability were changed. The observed relationship alone does not establish that answer. Differences in density, income, or existing travel needs might be related to both transit stops and car ownership.
Decision: The first question is about an observed relationship and supports an associational conclusion. The second asks about an effect of a change and needs evidence that supports causal interpretation. The wording changes the conclusion required, even though the variables are the same.
Prediction Is Not Explanation
Prediction and explanation can use the same variables, but success in one task does not guarantee success in the other. A forecast can be useful because a pattern repeats, even when the reason for that pattern is not understood. Conversely, a variable may be important for understanding a process but add little practical value to a particular forecast.
A simple thought experiment shows why predictive success does not prove a cause. Suppose prior scores are consistently useful for forecasting later scores. A forecast based on prior scores might be accurate because past and future performance tend to resemble one another. That accuracy does not show that changing the prior score itself would change the later score: the prior score is a record of earlier performance, not an intervention.
The logical distinction is straightforward. Prediction evaluates whether information helps anticipate outcomes. Causal explanation asks what would happen under a change to a factor. The first concerns forecast performance; the second concerns the effect of a change. Evidence for one statement is not automatically evidence for the other.
Worked Example: Predicting a Wait Time
Suppose a clinic has hypothetical records showing that, on busy days, patients who arrive later in the morning often wait longer. The clinic wants to use arrival time and the day’s appointment load to estimate a new patient’s wait.
The request is predictive: it asks for an expected wait for a case based on information available about that case. A prediction can be evaluated by comparing forecasts with the wait times later recorded for patients.
The clinic might also ask, “Does appointment load help explain variation in waiting time?” That is explanatory, because it asks about a relationship. Even if appointment load improves the forecasts, that result alone does not establish what would happen if the clinic changed its schedule. Prediction assesses whether a forecast is useful; the explanatory question concerns a relationship; a claim about the effect of a scheduling change goes further.
Decision: Classify the question from its requested conclusion. “Estimate a new patient’s wait” is predictive. “Is appointment load related to waiting time?” is explanatory. “What is the average wait recorded this month?” is descriptive.
When One Project Has More Than One Question
A project can include descriptive, explanatory, and predictive questions in sequence. For example, an analyst might first summarize recorded delivery times, then investigate whether distance is related to delivery time, and finally forecast a delivery time for a new order. The project has not become one kind of analysis throughout; it contains three distinct aims.
This is useful because data and techniques do not determine the aim on their own. A regression analysis might produce summaries of observed patterns, quantify a relationship, or generate forecasts. The intended use of the result decides which interpretation is appropriate. A clear research question keeps the conclusion focused and helps prevent overstatement.
Common Mistakes
- Calling every question about two variables predictive. Asking whether variables are related is explanatory; asking what outcome to expect for a case is predictive.
- Calling every relationship question causal. A relationship can be described without showing that one variable causes another. Use causal language only when the evidence supports it.
- Treating accurate forecasts as explanations. A forecast can work without identifying the process that produces the outcome.
- Classifying by the data rather than the aim. The same records can answer descriptive, explanatory, and predictive questions, depending on what is being asked.
- Reporting more than the question supports. Match the conclusion to the task: describe what is observed, characterize the relationship, or report a forecast.
Check Your Understanding
For each question, identify its main purpose and state what kind of conclusion it can support.
- A school asks, “What was the distribution of reading scores among the students assessed this term?”
- A researcher asks whether daily screen time is related to hours of sleep among a group of teenagers. Is this question descriptive, explanatory, or predictive?
- A coach uses a runner’s recent training records to estimate the runner’s finishing time in an upcoming race. What is the main purpose?
- A city finds that areas with more bike lanes have more bicycle trips. Why does this pattern alone not show that adding bike lanes caused the difference?
- A model forecasts next month’s sales accurately from recent sales. Does that accuracy alone explain why sales change? Explain briefly.