Survey Data Analysis using Power BI

Introduction
Recently, I analysed survey data from the research conducted to study the impact of fitness wearables on consumer behaviour, which can be found on Kaggle. The dataset was collected for beginners to perform exploratory data analysis. However, I used this data to practice data modelling, exploratory data analysis, and report/dashboard generation in Power BI.

The Dataset
The dataset consists of 30 responses from 30 respondents and 21 questions that were asked along with the timestamp. This data is tidy, i.e., it is complete with no missing values, every row corresponds to an observation, each column corresponds to a feature in each response, and each cell contains only a value.

Business Task
The three key business questions I identified from this data are

1. Who are those using the fitness wearables, i.e. consumer demographic?
2. What is the trend in the use of fitness wearables?
3. What is the behavioural impact of fitness wearables on consumers?

Due to the nature of the datasets, because it has too many columns with the same answer choice, it is crucial to organise and structure data in a way that makes it easier to understand and analyse.

Data Analysis
The analysis procedures meted on this dataset are data cleaning, modelling and data visualisation. The data cleaning process involved removing the timestamp column, adding a responseID column and renaming some columns.

Data Modelling.
After carefully observing the dataset to model the data so that its analysis will answer those business questions listed better, I realised that some survey questions relate to the impact of fitness wearables on consumers. These impacts are listed below;

• Helped them connect to the fitness community
• Helped achieve their fitness goal
• Impacted their overall health
• Improved overall well-being
• Improved sleep patterns
• Made exercising more enjoyable
• Stay motivated to exercise

I grouped all questions related to these and the responseID column (to link the table to other tables in the data model) in a table called impact and unpivoted them, with their answers ranging from strongly disagree to strongly agree

The other group of questions relates to the influence that fitness wearables have had on consumer behaviour, and they are;

• To change diet
• To exercise more
• To join a gym or fitness class
• To purchase other fitness-related products

Questions related to these and the ResponseID column were grouped and unpivoted in another table called influence.
Other questions related to the consumers' demographics and the trend of fitness wearables use were left in the survey table. Custom sort tables for impact and influence tables were created to sort the tables from Strongly Disagree being the least to Strongly agree being the highest ranking.
The final data model is shown below.

Data Visualisation

The overview of the visualisations is shown in the Power BI dashboard below. All of the features are categorical columns; as such, donut chart, column and stacked bar chart were used.

In the visual, we could see that the gender of respondents is almost evenly distributed and about 75% are age 34 and below. Only about 10% of the respondents rarely use fitness wearables. 80% of the respondents reported that fitness wearable has positively impacted their fitness routine.

All respondents agree that fitness wearable has influenced them to change their diet, exercise more and join a gym or fitness class. At the same time, only about 2% disagree that it has influenced them to purchase other fitness-related products.

A more significant percentage of the respondents agreed that it has strongly helped them connect to the fitness community, helped them achieve their fitness goal, impacted their overall health, improved overall well-being, improved sleep patterns, made exercising more enjoyable and stayed motivated to exercise.

Conclusion
Despite the small nature of the dataset, significant insights can be made. My data modelling approach gave me a better understanding of how insights can be made from a dataset. However, for a more robust analysis, more data is required.

Link to full Power Bi report (pdf version) here

Thanks for reading. Please share your thoughts in the comment section below.