Databases and quality checks

Demystifying Decision Science

Howard Friedman

Adjunct Professor at Columbia University

Triaging data sources

Not all data is equally useful

Some datasets require heavy cleaning or lack meaningful features
Others may not be worth the effort at all

Data triaging helps prioritize quickly

Assess whether a dataset is viable before investing time and resources

Start with availability

Is the data easy to access or behind a paywall
Are there permission or security barriers
Sorting this out early helps avoid wasted effort

Key checks

Consider costs

Are there licensing or access fees
Will you need to pay for storage or specialized tools

Evaluate utility

Does the dataset contain what you need for your analysis
Check the scope, detail, completeness, and feature relevance

Check update frequency

Real-time predictions need real-time or regularly updated data
Mismatched update cycles can render data unusable

Assess geographic resolution

Does the data align with the spatial level your analysis requires
National-level data won't help if you need zip-code granularity

Data quality checks

Data quality matters

Bad data can ruin models and lead to bad decisions
"Garbage in, garbage out" is a real risk in decision science

Watch for missingness

Percentage of missing values in each column
Patterns that suggest deeper data collection issues

Do basic range checks

Are there values that don't make sense for your context
Examples: age = 450 or price = -$10
Use checks to flag major issues early

More checks

Review outliers

Use histograms or boxplots to visualize
Discuss with stakeholders before excluding anything

Assess timeliness

Ask whether your data is recent enough to be relevant
Fast-moving contexts need frequently updated data

Ensure formatting consistency

Look for inconsistent date formats, spelling issues, or column naming mismatches
Clean and standardize to avoid problems during analysis

Let's practice!

Demystifying Decision Science

Preparing Video For Download...