ECONOMETRICS &
MARKETING ANALYSIS

Applied econometric modeling techniques in RStudio to evaluate how campaign engagement, advertising exposure, acquisition cost, campaign duration, campaign type, and channel mix influence digital marketing conversion rates.

Focus: Regression Analysis
Dataset: 200,000 Campaigns
Method: OLS + Robust SE
Tools: RStudio, R, ggplot2

PROJECT OVERVIEW

Background

This project analyzed a large-scale digital marketing campaign dataset to evaluate whether campaign engagement, advertising exposure, acquisition cost, duration, campaign type, and channel used meaningfully explain variation in conversion rates.

Core Objective

The objective was to apply econometric modeling and diagnostic testing to a marketing performance question, while interpreting the results through both statistical and practical business relevance.

RESEARCH QUESTION

Main Question

How do campaign engagement, exposure, and cost efficiency influence conversion rates across digital marketing campaigns?


  • Dependent Variable: Conversion Rate
  • Key Variables: Engagement Score, Clicks, Impressions, Acquisition Cost
  • Controls: Duration, Campaign Type, Channel Used
  • Data Type: Cross-sectional campaign-level data

TOOLS USED

RStudio R ggplot2 dplyr lmtest sandwich car stargazer OLS Regression Marketing Analytics

METHODOLOGY

Data Preparation

  • Cleaned Acquisition Cost by removing currency formatting.
  • Converted campaign Duration into numeric day values.
  • Converted Conversion Rate, Clicks, Impressions, and Engagement Score into numeric variables.
  • Set Email as the reference category for Campaign Type and Channel Used.
  • Validated that key variables had no missing values.

Modeling Approach

The analysis used Ordinary Least Squares regression across five progressively expanded specifications, moving from a bivariate model to a full baseline model with categorical controls.

MODEL SPECIFICATION

Full Baseline Model

ConversionRate = β0 + β1EngagementScore + β2Clicks + β3Impressions + β4AcquisitionCost + β5Duration + β6CampaignType + β7ChannelUsed + u

The full model estimated the relationship between conversion performance and campaign-level variables while controlling for campaign format and channel differences.

KEY METRICS

200K Campaign observations analyzed.
5 Regression specifications tested.
1.00 Approximate VIF values across regressors.
0.1687 Breusch-Pagan p-value.

DIAGNOSTIC TESTING

Heteroskedasticity

  • Breusch-Pagan test statistic: 18.902
  • p-value: 0.1687
  • No strong evidence of heteroskedasticity was detected.
  • HC1 robust standard errors were still applied for reliable inference.

Multicollinearity

  • All VIF/GVIF values were approximately 1.00.
  • No variables exceeded the common threshold of 5.
  • No variables needed to be dropped or combined.

Outlier Analysis

  • Zero outliers were detected using the |z| > 2 threshold.
  • Adjusted R² was unchanged with and without outlier removal.

REGRESSION RESULTS

Full Baseline Findings

The full model produced an adjusted R² of approximately -0.000028, indicating that the independent variables explained essentially none of the variation in Conversion Rate. This reflected the synthetic structure of the dataset rather than a modeling error.

Variable Actual Sign Significant? Practical Meaning
Clicks Positive No Near-zero effect
Impressions Negative No Near-zero effect
Engagement Score Negative No Near-zero effect
Acquisition Cost Positive No Near-zero effect
Campaign: Influencer Positive 10% only Not practically meaningful

ROBUSTNESS CHECKS

Model Validation

  • Removed Impressions to test whether Clicks was sensitive to exposure variable inclusion.
  • Removed categorical controls to test coefficient stability.
  • Restricted the sample to Email and Search campaigns only.

Robustness Conclusion

Results remained stable across all robustness checks. Coefficients stayed near zero and statistically insignificant, confirming that the findings were driven by the synthetic data-generating process rather than model specification errors.

BUSINESS INSIGHT

What This Project Demonstrated

The key takeaway was that a statistically correct model cannot produce meaningful business insight when the underlying dataset lacks real relationships between variables. The project reinforced the importance of data quality, diagnostic testing, model validation, robustness checks, and practical interpretation beyond p-values.

FULL ANALYSIS

Interactive RPubs Report

View the full econometrics analysis, including model specification, regression outputs, diagnostic testing, robust standard errors, and final interpretation.

Open Full Report →

INTERESTED IN
MORE WORK?

Explore more marketing, analytics, paid media, and ad operations projects.

View All Projects Contact Mickyas