Evaluating Happiness

Project Snapshot

Client: NA

Goal: Identify features that predict life satisfaction ratings

Impact: Identify a way to examine factors that influence satisfaction that can later extend to investigating issues of retention and turnover/churn/attrition

Turnaround: 3 weeks (May - June 2021)

Programming Language: Python

Analytical Methods: Descriptive analytics, logistic regression, tree-based models, UMAP dimensionality reduction

Tools: Jupyter notebook, pandas, numpy, scipy, matplotlib, seaborn, plotly, sklearn, umap, ipywidgets

My roles: Independent project - decided on topic; proposed data questions; retrieved and cleaned data; designed analytical approach and methodology; performed analyses; presented findings

Description

• Analyzed the relationship between happiness and values using data from Wave 6 (2010-2014) of the World Values Survey

• Compared the utility of a logistic regression classifier and a tree-based ensemble learning algorithm (adaptive boosting) for accurately predicting happiness ratings based on values data

• Used feature importance scores to see which survey items were most useful for predicting happiness

• Employed UMAP dimensionality reduction to determine and visualize how survey respondents’ answers within and across countries compare

Key Findings

Moderate Correlates with Happiness

• For the most part, people are pretty happy (85%) and stay happy.

• Unsurprisingly, respondents who were in good health, felt that they had more freedom of choice, who were satisfied with their household income, and who belonged in a higher income group were typically less unhappy.

• Regardless of health or socioeconomic circumstances though, most respondents tended to rate themselves as happy more often than unhappy.

• These features seemed to reveal more about correlates with unhappiness than happiness.

• For the most part, people are pretty happy and stay happy.

• Interestingly, there were still high proportions of respondents in poorer health, who had less perceived freedom of choice, who were less satisfied with their household incomes, and who belonged to lower-income groups in their societies who also tended to rate themselves as happy more often than unhappy.

• These features seemed to reveal more about correlates with unhappiness than happiness.

Three Categories of Values Used to Classify Happiness

Based on a combination of expressed values (what people find important and/or identify as ideal) and respondent’s recorded experiences (with poverty, safety, health, etc.), both models tended to identify three types of values that are most useful for influencing the likelihood that a respondent will be classified as unhappy or happy by either model:

• Financial and physical security values/conditions

• Companionship values/conditions

• Freedom of choice and opportunity values/conditions

Comparing and Contrasting Survey Responses

UMAP analyses allowed me to explore whether random samples of respondents from the same country were responding to survey items in the same way. In the 2D representation below, each point represents one respondent. Note that country labels were assigned after clusters were formed.

Tight clusters of points signify similarity in survey responses, personal experiences, and viewpoints. Points of the same color that are spread out represent variety in survey responses and thinking.

Git Repository

For more details and to see my code, check out this project's repo