Why Define the Outcome Precisely?
A regression question needs a clearly defined outcome: the quantity the analysis is intended to describe, explain, or predict. Naming a broad concept is not enough. “Health,” “school performance,” or “treatment success” could each be measured in several ways, and those measurements need not lead to the same answer. A useful outcome definition tells a reader exactly what value is recorded, in what units, at what time, and which values count as eligible.
For example, asking whether a program is associated with better health does not yet identify an outcome. The outcome could be a symptom score at a specified follow-up, a change in that score from baseline, or whether a participant meets a defined threshold. These choices may be reasonable for different purposes, but they are not interchangeable. Defining the outcome is part of making the research question precise, not a clerical step to postpone until after data collection.
Here, “unit” can mean the measurement unit, such as kilograms, minutes, or points. The outcome definition should also make clear what one recorded value refers to—for example, a person’s score or a school’s annual rate. These details prevent a reader from mistaking one quantity for another. Broader questions about which cases are included in a study are separate design decisions.
Four Parts of an Outcome Definition
Start by naming the measured quantity in concrete terms. “Reading ability” is a construct or broad idea; “number of words read correctly in one minute on a specified test” is a more operational outcome. The definition should identify the measurement rule closely enough that two trained people would know what to record.
Next state the unit or scale. A value of 12 has little meaning without knowing whether it represents 12 points, 12 minutes, or 12 events. If the outcome is a score with no physical unit, name the score scale and, where relevant, its possible range. For categorical outcomes, list the categories and explain how each is assigned.
The time reference matters just as much. “Blood pressure” could mean the measurement at enrollment, the measurement after a specified follow-up, or an average across several visits. If a question concerns a period, define that period and the rule for summarizing it. Finally, state which values are valid. A recorded value outside a scale’s allowed range may be an error, while an unrecorded score is missing; neither should silently be treated as an ordinary value.
Alternative Definitions for the Same Research Problem
Suppose the broad research problem is to describe how participants’ body weight changes during a program. One possible outcome is weight at the end of the program. Another is the difference between weight at the end and weight at the start. The endpoint asks about weight at a specified time; the change outcome asks about the amount of increase or decrease over the interval. Both are measured in kilograms, but they answer different questions.
For a change outcome, state the subtraction order. If change is defined as follow-up minus baseline, negative values represent decreases and positive values represent increases. Reversing the order would reverse those interpretations. The measurement interval also belongs in the definition: a change over eight weeks is not the same quantity as a change over six months.
Worked Example: Endpoint Weight or Weight Change?
Hypothetical data. Three adults have weights recorded at the start of a program and again eight weeks later. Consider two candidate outcomes: weight at week 8, and change from baseline to week 8 defined as week 8 weight minus baseline weight.
| Participant | Baseline weight (kg) | Week 8 weight (kg) | Change (kg) |
|---|---|---|---|
| A | 70 | 68 | 68 − 70 = −2 |
| B | 82 | 80 | 80 − 82 = −2 |
| C | 65 | 66 | 66 − 65 = 1 |
For the endpoint definition, the mean week 8 weight is
For the change definition, the mean change is
The first outcome is weight at week 8, measured in kilograms; the second is change over eight weeks, also measured in kilograms. The mean change of \(-1.00\) kg indicates an average decrease of 1.00 kg in these three hypothetical records. It does not say that every participant lost weight: participant C gained 1 kg. These summaries describe the recorded values and, by themselves, do not establish why the changes occurred.
The relationship between these two definitions follows directly from subtraction. For each participant \(i\), define \(D_i=Y_{i,\text{follow-up}}-Y_{i,\text{baseline}}\). Averaging the changes gives
The equality holds because summing the differences is the same as subtracting the two sums, and both sums are divided by the same \(n\). In the example, the baseline mean is \((70+82+65)/3=72.33\) kg and the follow-up mean is \(71.33\) kg, so their difference is \(71.33-72.33=-1.00\) kg, matching the mean of the individual changes. This arithmetic does not make endpoint and change the same outcome; it shows how their averages are related when the same participants have both measurements.
Rates, Scores, and Thresholds
Other outcome choices may involve a denominator or a rule for converting a measurement into a category. A rate or proportion must name both its numerator and denominator and define the period over which they are counted. A score outcome should identify its scale and valid range. A threshold outcome must state the threshold and how equality is handled. Each definition retains some information and discards other information, so the choice should follow the question rather than convenience alone.
Worked Example: Defining an Attendance Outcome
Hypothetical data. A course organizer wants to describe attendance during each student’s scheduled course period. Student A was scheduled for four sessions and attended three; student B was scheduled for four and attended two; student C joined later and was scheduled for two, attending both. Define each student’s outcome as the percentage of that student’s scheduled sessions attended.
The measurement rule is
The three outcomes are
If the intended outcome is the mean of the three students’ attendance percentages, it is \((75+50+100)/3=75.0\%\). If instead the question concerns the combined fraction of all scheduled sessions attended, the calculation is \(100\times(3+2+2)/(4+4+2)=70.0\%\). These are different summaries because one gives each student equal weight, while the other counts each scheduled session equally. The outcome definition should say which quantity is intended; “attendance” alone does not resolve that choice.
The attendance example also shows why denominators need explicit rules. Counting every session in a fixed calendar period would treat the late-arriving student differently from counting only sessions scheduled after that student joined. Neither denominator should be chosen silently. The correct definition depends on the question being asked and should be stated so the resulting percentage has an unambiguous meaning.
Worked Example: Valid Scores and a Threshold Outcome
Hypothetical data. Five people complete a symptom questionnaire at a specified follow-up. The scale ranges from 0 to 10 points, inclusive, with larger scores representing more reported symptoms. The recorded values are 6, 3, 8, 11, and one missing response. Consider both the numeric score and a binary outcome defined as a valid score of at least 7 points.
The eligible numeric values satisfy \(0\leq Y\leq 10\). The value 11 is outside the scale’s stated range, so it is not a valid score under this definition. A missing response is also not a score; it should not be assigned the value zero. The three valid observed scores are 6, 3, and 8, with mean
For the threshold outcome, define \(Z=1\) when the valid score is at least 7 and \(Z=0\) when it is below 7. Among the three valid scores, only 8 meets the threshold, so the observed fraction meeting it is \(1/3=0.333\), or 33.3%. The numeric score retains distinctions among valid scores; the binary definition keeps only whether the threshold was reached. These outcomes address related but distinct questions.
Writing and Checking an Outcome Definition
A practical outcome statement can be written as a sentence: “The outcome is [quantity] measured in [units or scale] at [time or period], with valid values [range or categories] and [rule for derived values, if applicable].” If the outcome is derived from multiple measurements, include the calculation. If values are categorical, specify the category assignment rule. If a record is missing or invalid, state how it will be recognized rather than turning it into a substantive outcome value.
Replace broad labels such as “success” with the measurement or rule that produces a recorded value.
Give the physical unit, score scale, or categories, and identify the measurement time or period.
Specify allowable values, category rules, denominators, or thresholds, as appropriate.
Explain what each candidate outcome represents and what distinctions it preserves or removes.
Common Mistakes and Why the Choice Matters
- Leaving timing vague. “Score” could mean baseline, follow-up, or an average across visits. Name the time point or period.
- Reporting a number without its scale. A value of 8 is not interpretable as a score unless the scale and meaning of larger values are stated.
- Changing a definition without noticing. Follow-up level, change from baseline, and reaching a threshold are different outcomes, even when based on the same measurements.
- Treating missing or invalid values as real values. A missing score is not zero, and a score outside the allowed range should not be silently accepted as valid.
- Leaving a denominator implicit. A percentage is ambiguous unless the counted events and eligible opportunities are defined.
A common mistake is to say, “The outcome is improvement, so we will use the follow-up score.” That reasoning assumes that the follow-up score directly measures improvement. It may instead describe the final level; participants with different starting scores can have the same follow-up value but different changes. The correction is to state the intended quantity explicitly: use a follow-up level if the question concerns level at follow-up, or define improvement as a particular change or threshold rule if the question concerns improvement. The chosen outcome should match the wording of the research question.
Check Your Understanding
For each question, identify what additional information would make the outcome definition unambiguous.
- A team wants to study students’ “academic performance.” Give one precise score-based outcome definition, including timing and an eligible range or scale.
- For a question about change in resting heart rate over 12 weeks, state the subtraction order, units, and the interpretation of a negative change.
- A report gives an attendance percentage. What must be specified about the numerator, denominator, and measurement period?
- A 0-to-20 score has one recorded value of 24 and one blank entry. Explain how each differs from a valid score of zero.
- Give two distinct outcome definitions for a question about participants’ symptoms, and state briefly what question each definition answers.