Week 3 – Data Cleaning, Integration and Exploration of NHS ERIC Data

03 Jul 2026 - Sneha Dharmeche - climate risk, data preprocessing, healthcare

This week focused on progressing from data familiarisation into more structured data cleaning, integration and exploratory analysis using the NHS Estates Returns Information Collection (ERIC) dataset.

Data Cleaning and Preparation

A key focus this week was completing more detailed data cleaning across multiple yearly datasets (2021–2025). Using OpenRefine, I standardised column names, ensured numerical variables were correctly formatted and removed inconsistencies such as extra spaces in categorical fields.

I also carried out a review of missing values across all variables to understand data quality issues. This included distinguishing between true missing values, zero values and “Not Applicable” entries.

Through this process, I found that most variables are relatively complete. However, certain fields such as backlog maintenance costs and overheating occurrences contain substantial missingness. Initial investigation suggests that this missingness is not random and is likely linked to site characteristics such as support facilities, unoccupied sites or leased/PFI arrangements where certain metrics are not applicable.

Data Integration & Visualisation

After preparing the combined dataset, I began initial exploratory data analysis (EDA) to better understand trends in key variables.

Visualisations I Did:

Total flooding incidents over time
Overheating-related risk assessment events over time
Missingness patterns across years

Challenges and Reflections

A key challenge this week was interpreting missing data appropriately, particularly distinguishing between missing values due to data quality issues and those due to structural “Not Applicable” cases. This required revisiting documentation and carefully examining how different site types are represented in the dataset.

Another challenge was ensuring consistency across multiple yearly datasets, particularly as variable definitions and site coverage are not perfectly uniform across all years.

Despite these challenges, , this week felt like a clear step forward from just cleaning the data into actually starting to explore it. I now have a better understanding of how the dataset is structured, which will help guide the next stages of my work.