The proxy trap: When easy data measures the wrong thing
You want to improve a practice, but the easiest available data show whether someone opened a module or checked a box. Those data may be useful. The problem begins when you treat them as proof of something they do not directly measure.
A proxy is an indirect indicator used to stand in for a concept. The key question is not whether proxies are always wrong. It is whether the evidence supports the interpretation you want to make. SQUIRE asks you to explain why a measure was chosen and how its validity was assessed. [1]
Follow the concept, not the available field
Hypothetical example. A clinic wants patients to understand appointment preparation instructions. The electronic record contains a field labeled “instructions provided.” After staff training, that field is completed more often.
The change supports a conclusion about documentation. It may reflect more instruction, better recording of existing instruction, or both. It does not by itself show that patients understood or followed the instructions.
| Intended concept | Convenient indicator | What that indicator directly supports |
|---|---|---|
| Patient understanding | “Instructions provided” checkbox | Documentation that instruction was recorded. |
| Staff competence | Self-rated confidence | How capable staff say they feel. |
| Skill acquisition | Training attendance | Exposure to the training opportunity. |
| Behavior in practice | Knowledge quiz score | Performance on the quiz, not routine behavior. |
These measures can belong in a project when they match a stated question. A confidence outcome is not wrong simply because it is not a competence measure. It becomes misleading when the project claims competence from confidence alone.
Choose a closer measure when possible
For the instruction project, you could assess whether patients identify the correct preparation steps using a consistent brief check. That moves the measurement closer to understanding. It still does not prove that they will follow the instructions later.
You might retain the documentation field as a process measure and use the understanding check as a nearer outcome. This gives each measure a clear role rather than asking one convenient field to answer every question.
A closer measure may require more staff time, training, or permission. Balance those demands openly. When a proxy is the only feasible choice, name it accurately, explain the reason for using it, and limit the conclusion to what it can support.
Your next decision
Write two sentences: “I want to know whether ___ changed” and “My data directly show ___.” If they name different things, identify the missing link. Either strengthen the measurement or narrow the claim. See Valid, feasible, relevant: Choosing outcome measures for selecting measures and Outcome, process, and balancing measures for assigning their roles.