Why Record Change Within Each Subject?
In a before-and-after study, the same subjects are measured at two times: once at baseline and again at follow-up. The measurements are linked because each follow-up value belongs to the same subject as a particular baseline value. Rather than treating the two sets of measurements as unrelated, we can calculate how much each subject’s measured value changed.
This calculation gives a compact description of the difference between two measurements on one subject. It can reveal whether the observed values tended to rise or fall and by how much, using the original measurement units. But it describes change between measurement occasions; by itself, it does not tell us why the change occurred.
This tutorial uses follow-up minus baseline throughout. Under that convention, a positive value means the measurement is higher at follow-up, a negative value means it is lower, and zero means the two recorded values are equal. Whether a higher value is desirable depends on the variable being measured: an increase in a knowledge score may be favourable, while an increase in a symptom score may not be.
Calculating and Reading a Change Score
The calculation is performed separately for each subject. The pairing is essential: subtract the baseline value for a subject from that same subject’s follow-up value, not from a different subject’s measurement. A table makes the correspondence visible.
Worked Example: Change in a Clinical Measurement
In a hypothetical clinic project, five adults have their systolic blood pressure recorded at baseline and again four weeks later. The values below are illustrative, and the project description does not establish that any particular intervention caused the changes.
| Subject | Baseline (mmHg) | Follow-up (mmHg) | Change: follow-up − baseline (mmHg) |
|---|---|---|---|
| A | 150 | 142 | −8 |
| B | 142 | 138 | −4 |
| C | 160 | 151 | −9 |
| D | 155 | 149 | −6 |
| E | 148 | 141 | −7 |
For subject A, the calculation is \(142 - 150 = -8\) mmHg. The negative sign indicates that this subject’s recorded systolic blood pressure was 8 mmHg lower at follow-up. The other changes are calculated in the same way.
The arithmetic mean of these five observed changes is
Thus, in this hypothetical set of five complete pairs, the average observed change was −6.8 mmHg: the follow-up values were, on average, 6.8 mmHg lower than the baseline values. This is a description of these recorded measurements, not proof that a treatment produced the decrease.
The change score retains the direction and size of a difference in the variable’s original units. That makes the result interpretable in context: −6.8 mmHg is a difference in blood pressure, not a percentage or a unitless score. The score also compresses the two measurements for a subject into one number, so it can answer questions such as, “How much did this subject’s recorded value change between these occasions?”
What a Change Score Measures
A change score measures the net difference between two recorded values. It does not describe every event between the measurement occasions, and it does not reveal the path taken from baseline to follow-up. For example, a score of zero means the two recorded endpoint values match; it does not mean the subject’s value stayed constant at every moment in between.
The subtraction also makes clear which part of the data is being compared. If a subject’s baseline value is \(Y_{i0}\) and follow-up value is \(Y_{i1}\), then the change is the amount that must be added to the baseline value to obtain the follow-up value:
This identity is a useful check on the arithmetic. It also explains why change is personal to the subject: the baseline and follow-up values are combined within the same subject’s record. Differences in starting levels between subjects are not themselves counted as changes.
For instance, two subjects can end at the same follow-up value but have different change scores because they started at different levels. Conversely, two subjects can have the same change even if one has much higher measurements at both occasions. A change score describes movement between the two recorded values, not the subject’s absolute level at either time.
Worked Example: Change in a Knowledge Score
In a hypothetical training evaluation, four participants take the same 20-point knowledge assessment before and after a short course. The score ranges from 0 to 20, with higher scores indicating more correct answers. The recorded values are:
| Participant | Baseline score | Follow-up score | Change (points) |
|---|---|---|---|
| A | 8 | 12 | 4 |
| B | 14 | 16 | 2 |
| C | 10 | 9 | −1 |
| D | 6 | 11 | 5 |
For participant C, \(D_C = 9 - 10 = -1\) point, a one-point decrease. For participant D, \(D_D = 11 - 6 = 5\) points, a five-point increase. A positive change means a higher score at follow-up, and, given the score’s definition, more correct answers.
The mean observed change is
The four participants’ scores increased by an average of 2.5 points between the two assessments. This summary does not mean every participant improved: participant C’s score decreased. Nor does it establish that the course caused the average increase; practice with the assessment or other events could also contribute.
Change Does Not Identify Its Cause
A before-and-after record establishes the order of the two measurements, but an observed difference is not automatically an effect caused by a particular intervention. Other influences may change between baseline and follow-up, including ordinary time trends, changes in behaviour, additional care, differences in measurement conditions, or random variation in the recorded values.
A simple decomposition illustrates the limitation. Suppose a measurement for subject \(i\) at time \(t\) can be represented as
where \(b_i\) is a component stable for that subject across the two occasions, \(g_{it}\) represents components that may vary over time, and \(e_{it}\) represents measurement error or other short-term variation. Subtracting baseline from follow-up gives
The stable component \(b_i\) cancels in this representation. However, the change still includes all components that differ between the two occasions, not just any intervention of interest. The decomposition is a way to understand the subtraction; it does not guarantee that real measurements follow this model exactly or that the intervention can be separated from other changes.
Worked Example: A Decrease in a Symptom Score
In a hypothetical follow-up program, a symptom scale runs from 0 to 30, with higher scores indicating more severe symptoms. One participant scores 18 at baseline and 13 at follow-up. Calculate and interpret the change.
Using the stated convention,
The score was 5 points lower at follow-up. Because higher values indicate more severe symptoms, this is a decrease in measured symptom severity for this participant. The negative sign does not mean the change was necessarily harmful; its practical direction depends on what higher scores mean.
It would be inaccurate to conclude from this pair alone that the program caused a five-point improvement. The calculation contains no comparison with what would have happened to this participant over the same period without the program. The defensible description is that the recorded symptom score decreased by 5 points between baseline and follow-up.
Keep the Measurement and Timing in View
A change score is only as interpretable as the two measurements used to create it. The variable should mean the same thing at baseline and follow-up, and the units and measurement procedures should be comparable. If a scale, device, observer, or measurement protocol changes, some apparent change may reflect the changed measurement process rather than a change in the subject.
The time interval also matters. A change from baseline to one week later answers a different descriptive question from a change from baseline to one year later. Reports should identify the measurement occasions or the elapsed interval, rather than describing a change without saying when it occurred.
A raw change is not the same as a percentage change. For example, an increase from 10 to 12 units is a raw change of 2 units. Relative to the baseline value, it is a 20% increase, calculated as \((12 - 10)/10 \times 100\%\). The percentage uses the baseline as a denominator and answers a different question; it can also be difficult to interpret when baseline values are zero or near zero. Unless a relative change is specifically needed, keep the raw difference and its original units clear.
Common Mistakes
- Reversing the subtraction without saying so. With follow-up minus baseline, an increase produces a positive score and a decrease produces a negative score. State the order used and apply it consistently.
- Subtracting measurements from different subjects. A change score requires the baseline and follow-up values for the same subject. Pairing the wrong records creates a number that does not describe either subject’s change.
- Reading the sign without knowing the scale. Positive means higher at follow-up, not automatically better; negative means lower, not automatically worse. Use the meaning of the variable to interpret direction.
- Calling observed change a treatment effect. Before-and-after subtraction describes the difference between recorded occasions. Other events or measurement variation may also explain some or all of it.
- Confusing raw change with percent change. A difference in the original units and a percentage relative to baseline are distinct summaries. Identify which is reported.
Check Your Understanding
Answer each question using follow-up minus baseline unless the question states otherwise.
- A participant’s baseline value is 24 units and follow-up value is 19 units. What is the change, including its sign and units?
- A scale measures fatigue from 0 to 40, with higher scores indicating more fatigue. What does a change of \(+3\) points mean, and does it necessarily indicate improvement?
- Why must a subject’s baseline value be paired with that same subject’s follow-up value?
- A subject’s measurement is 7 units at both baseline and follow-up. What can be concluded from a change score of zero, and what cannot be concluded about the time between measurements?
- In a before-and-after study, the mean observed change is negative. Give one reason this does not, on its own, prove that the intervention caused a decrease.