Confounding Adjustment Disclosure

A confounder is a variable that is associated with both the predictor and the outcome —and is therefore capable of creating an apparent relationship between them that is not causal. Confounding is the central threat to validity in observational research and a genuine threat even in experiments with imperfect randomization. Its detection requires declared metadata; its adjustment requires an explicit model choice.

STEP 1 —The Pitfall

The most common scenario: a researcher measures the effect of treatment A vs. treatment B on an outcome. Age is also recorded. Age is associated with the treatment assignment (older patients more likely to receive B) and with the outcome (older patients have worse outcomes). Without adjusting for age, the apparent treatment effect is confounded —some of the "treatment B effect" is actually an age effect.

The researcher may know this and adjust for age. Or they may not check, and report an unadjusted estimate as if it represents the causal effect of treatment. The difference between adjusted and unadjusted estimates can be large —large enough to change the direction of the effect.

[Image placeholder: DAG (directed acyclic graph) showing Age xTreatment and Age xOutcome, creating a backdoor path from Treatment xOutcome through Age. Before adjustment: apparent effect = 1.8x After adjustment for Age: true effect = 1.1x The confounding inflated the effect by 64%.]

Directed acyclic graph (DAG) illustrating confounding. Age opens a backdoor path from treatment to outcome. Adjustment closes the path and reveals the true estimate.

STEP 2 —Journal Requirement

STROBE (observational studies) requires that all analyses report both unadjusted and adjusted estimates. CONSORT requires disclosure of any variables that were adjusted for in the primary analysis and the basis for their inclusion. Epidemiology journals require a directed acyclic graph (DAG) or at minimum an explicit statement of the confounders identified and the adjustment strategy.

STEP 3 — What Licklider Currently Provides

What the current product does

The current product can record disclosure and acknowledgment related to confounding risk. Batch and plate confounding risk is surfaced via the batchConfounding gate and Card 36 (see Batch / Plate Confounding). The assumption ledger and export-facing state can reflect whether confounding-related disclosures have been made.

What is described here as methodology guidance

The STEP 1 and STEP 2 sections above describe general methodology for identifying and adjusting for confounders. These are standard practices, not claims about current product automation.

What the product does not do

The product does not automatically screen covariates for association with predictor and outcome.
The product does not compute correlation thresholds or flag potential confounders based on |r| or p-value thresholds.
The product does not display unadjusted and adjusted estimates in parallel.
The product does not auto-suggest DAGs or confounder declarations.
Automatic covariate adjustment execution is not part of the current product.

Known limitations

Current product behavior covers disclosure and acknowledgment of confounding risk. The researcher is responsible for identifying confounders, choosing adjustment strategies, and verifying that reported estimates reflect appropriate adjustment.

[Image placeholder — not current product UI. Illustrative concept of a confounding adjustment panel.]

Illustrative concept only. The current product records confounding-related disclosure state but does not expose this specific panel layout.

STEP 4 — Draft Output (Draft / Needs review)

The following is an example of how a researcher might write a confounding adjustment disclosure, not output that Licklider generates automatically.

The primary analysis adjusted for age and baseline disease severity,
identified a priori as potential confounders based on their known
associations with both treatment assignment and the primary outcome.
Unadjusted estimates are provided in Supplementary Table 5 for reference.
We acknowledge that residual confounding from unmeasured variables
(e.g., socioeconomic status, comorbidities) cannot be excluded.