Instrumental Variables (IV)
Uses an external 'instrument' that affects treatment but not outcome directly to identify causal effects.
Last updated
Definition
IV solves the endogeneity problem: when treated and untreated groups self-select (e.g., engaged users see your ad), naive comparison overstates the effect. A valid instrument is correlated with treatment but uncorrelated with the outcome's confounders. Hard to find in marketing — common instruments are weather, exogenous policy shocks, or experimental random rollouts.
How it applies in India
In Indian marketing data, a change that hits only some PIN codes, such as a new delivery fee, can serve as an instrument only if it moves sales through no route other than the one being studied, which is rarely easy to show.
An example
A skincare brand raises its Meta budget every festive season, exactly when shoppers were going to buy more anyway, so comparing high-spend and low-spend weeks overstates what the ads did. Suppose the ad auction became much more expensive for one week because of an unrelated event, so the brand bought fewer impressions than it planned. That price shock changed how many people saw the ads without changing how much those people wanted to buy, so it can act as an instrument: the dip in sales that follows the forced cut in exposure estimates the effect of the ads.
How WatEase uses it
WatEase does not fit instrumental-variable models. It answers the same question, what did the marketing cause, with randomised holdouts instead: shoppers held back from a recovery drip campaign or from a broadcast launched by an AI agent, and towns held back from live ads for 14 days in the quarterly geo-lift test. Randomising who is held back removes the self-selection problem an instrument exists to solve, without waiting for a natural experiment to come along.
Frequently asked questions
Does WatEase use instrumental variables to measure ads?
No. WatEase measures lift with holdouts: a share of shoppers or towns is kept out of a campaign and compared with everyone else, and the result is shown with a figure for how sure the comparison is. A randomised holdout removes the same bias an instrument is used for.
Related terms
Try WatEase
Run your WhatsApp commerce on the platform built for India — Cloud API, GST invoices, UPI checkout, opt-in tracking, and 0% markup on WhatsApp messages (competitors add 20–25%). Pricing is on request — quoted by the WatEase team.
Get Started Today