Paid search modeling

Understanding query volume as a confounder for search ads

Perhaps the biggest challenge in causal inference when applied to marketing is that advertisers often spend more on marketing when there is stronger demand for their product. Disentangling whether an increase in the KPI is due to an increase in marketing spend or due to an increase in inherent demand is a primary concern when one is analyzing causal effects of marketing spend.

The strong relationship between inherent demand and marketing spend is particularly salient when it comes to search ads. This is because a search ad is only shown on the page if a search query matches certain keywords targeted by a set of advertisers. When inherent demand is high, organic query volume will also be high, and so the total spending on search ads will be high. Therefore, organic query volume is an important confounder for search ads. It is hard to get good inference on search ads without it.

This is particularly an issue for advertisers with high search budgets because their paid search ad volume tends to track more closely with organic query volume. However, this also affects lower budget advertisers who increase their budgets during periods of high demand, or who only run search campaigns during these periods.

Meridian provides the option to include Google organic query volume (GQV) in the model as a confounder for Google Search ads. Organic query volume from non-Google search engines is often unavailable. If you want to model non-Google paid search ads, and organic query volume from the corresponding search engine is not available, the following alternatives might work for you:

  • Bias can be mitigated if GQV is a good proxy for the non-Google query volume. We recommend assessing this assumption. One way to help assess the assumption is by creating a plot, for example:

    Correlation between media impressions and
GQV

    The previous plot shows the correlation between media impressions and brand GQV on the y-axis, and the correlation between media impressions and generic query volume on the x-axis.

  • If you don't want to assume GQV is a good proxy for the non-Google query volume, you might need to omit the non-Google search engine from the model.

For more information about the challenges of selection bias due to ad targeting, see Bias Correction For Paid Search In Media Mix Modeling.