Why does my MMM output look wrong?

If channel spends always moved together, contribution cannot be separated.

Does high VIF mean I cannot run MMM?

You can run it, but per-channel separation will not be trustworthy. Interpreting grouped channels, or re-estimating after varying spend patterns, is the accurate path.

Why is a coefficient negative?

Collinear inputs make estimates unstable and signs can flip. Such a negative should be read as a symptom of non-identification, not as evidence of cannibalization.

How do I create unique variation?

Stagger increases and decreases across channels, or run a channel only in selected periods or regions, so spend schedules stop moving in lockstep.

Channel overlap

VIF multicollinearity check

Before MMM, check whether channel spend moves too tightly together to separate contribution.

Summary

Check whether channel spend moved too tightly together before running MMM. This is a diagnostic signal, not a causal fix.

Reading VIF

Below 5 is observationally OK; 5+ is a warning; 10+ or perfect collinearity (∞) is severe. Insufficient observations or variation leave the diagnosis uncomputed and held separately. Date, channel, and spend are required.

Before MMM, quickly check VIF to see whether channel spend moved too tightly together to separate contribution.Needs: Date · Channel/campaign · Ad spend3 required columns

Loading…

Next step

Channel contribution

Choose after the verdictPrepare a new dataset

Channel contribution (MMM)

Find other analyses