Digital Marketing Campaign Performance

Reference Categories:

• Campaign Type → Email (baseline)
• Channel Used → Email (baseline)

All dummy coefficients interpreted relative to Email.

Estimation: OLS with heteroscedasticity-robust standard errors (HC1) applied to all specifications.

What the data shows

• Influencer campaigns: +0.000526 above Email baseline
• Marginally significant at 10% (p = 0.067)
• All other campaign types: not significant

Why likely a statistical artifact

With n = 200,000, even trivially small differences become detectable. The magnitude (+0.053% above baseline) has no practical marketing relevance. The synthetic data-generating process likely introduced a small random difference that the large sample inflates into marginal significance.

Could it be theoretically real?

There is a theoretical reason to expect influencer campaigns to outperform email:

• Influencer content is perceived as authentic — reducing ad skepticism
• Email requires opt-in behavior and faces inbox competition
• Influencers reach high-intent niche audiences that email lists may miss

Our result is directionally consistent with theory — even if not meaningful in magnitude on synthetic data.

What the null results mean

These are legitimate findings, not a modeling failure. A well-specified model on synthetic data correctly returns null results when variables were generated independently.

“Null results from a well-specified model on a synthetic dataset are an honest finding.”

Threats to Internal Validity

1. Omitted Variable Bias — Unmeasured factors like brand awareness or audience targeting quality likely affect conversion rate and correlate with our regressors. Partially addressed by including Engagement Score and Acquisition Cost as proxies.
2. Functional Form Misspecification — A linear model is appropriate given the uniform distribution of Conversion Rate, but real data may require log or quadratic transformations of Clicks and Impressions.
3. Measurement Error — Engagement Score construction is undocumented; if it is a noisy proxy for true engagement, its coefficient would be attenuated toward zero.

Threats to External Validity

4. Sample Selection — Data is synthetic and variables are independently generated with no real-world covariance structure. Findings cannot generalize to actual marketing campaigns.
5. Simultaneous Causality — In real data, conversion rate could affect future ad spend and campaign type choices, creating reverse causality. Cross-sectional OLS cannot address this; IV or experimental designs would be required.

This project as a template

Despite these limitations, our specification, cleaning pipeline, diagnostics, and robustness checks form a fully reproducible framework ready to apply to real marketing data — where theoretically predicted relationships should emerge.

Main Findings

1. No variable significantly explains Conversion Rate — all coefficients near zero, Adj R² ≈ 0
2. Results consistent across all robustness checks — dropping variables or restricting the sample does not change the core finding
3. Influencer dummy marginally significant at 10% — directionally consistent with theory, but likely a large-sample artifact on synthetic data
4. No econometric issues detected — BP test homoscedastic, all VIF ≈ 1.0, zero outliers

Actionable Recommendation

Firms should not assume that channel type or campaign format alone drives conversion rate. Before optimizing campaign mix, invest in real tracking infrastructure and data quality — specifically, linking ad exposure to individual-level conversion outcomes. Until that covariance structure exists in the data, any model will return null results regardless of how well-specified it is. This framework is ready to deploy the moment that data is available.

If We Could Continue

• Apply to real campaign-level data
• Explore log transformations of Clicks and Impressions
• Test interaction terms (e.g., Engagement Score × Campaign Type)
• Address endogeneity via IV or natural experiment

Project Contributions

Academic Sources

• Manchanda, P., Dube, J. P., Goh, K. Y., & Chintagunta, P. K. (2006). The effect of banner advertising on internet purchasing. Journal of Marketing Research, 43(1), 98–108.

• Kireyev, P., Pauwels, K., & Gupta, S. (2016). Do display ads influence search? Attribution and dynamics in online advertising. International Journal of Research in Marketing, 33(3), 475–490.

Academic Sources (cont.)

• Blake, T., Nosko, C., & Tadelis, S. (2015). Consumer heterogeneity and paid search effectiveness: A large-scale field experiment. Econometrica, 83(1), 155–174.

• Lambrecht, A., & Tucker, C. (2013). When does retargeting work? Information specificity in online advertising. Journal of Marketing Research, 50(5), 561–576.