Survey Stats

This skill provides a suite of statistical tools optimized for market research and survey analysis. It covers the end-to-end workflow from weighting sample data to extracting deep insights via multivariate analysis.

Core Capabilities

1. Sample Weighting (weighting.py)

Adjusts your sample to match population targets using Iterative Proportional Fitting (Raking).

Usage:

from scripts.weighting import rake_weights

targets = {
    'gender': {'Male': 0.49, 'Female': 0.51},
    'region': {'North': 0.2, 'South': 0.8}
}
df['final_weight'] = rake_weights(df, targets)

2. Factor Analysis (factor_analysis.py)

Reduces large sets of variables (e.g., Likert scales) into latent factors. Useful for identifying underlying themes in "Importance" or "Satisfaction" batteries.

Usage:

from scripts.factor_analysis import run_factor_analysis

# Select numerical columns (Likert scale 1-5)
cols = ['q1_sat', 'q2_sat', 'q3_sat', ...]
loadings, variance = run_factor_analysis(df, cols)

3. Clustering / Segmentation (clustering.py)

Groups respondents into homogeneous segments based on behavior or attitudes using K-Means.

Usage:

from scripts.clustering import run_segmentation

cols_to_segment = ['factor1_score', 'factor2_score']
df_segmented = run_segmentation(df, cols_to_segment, n_clusters=4)

4. TURF Analysis (turf_analysis.py)

Calculates the Total Unduplicated Reach and Frequency. Perfect for optimizing product lines or communication channels.

Usage:

from scripts.turf_analysis import run_turf_analysis

items = ['Channel_A', 'Channel_B', 'Channel_C']
turf_res = run_turf_analysis(df, items, n_max_size=3)
print(turf_res)

5. Survey PCA (survey_pca.py)

Principal Component Analysis tailored for survey data, including support for standard Likert scales AND multi-response (string-split) data.

Usage:

from scripts.survey_pca import run_survey_pca

# For multi-response columns like "ItemA;ItemB"
cols = ['multi_resp_q1', 'multi_resp_q2']
loadings, scores, var = run_survey_pca(df, cols, sep=';')

6. Weighted Visuals (visuals.py)

Helper functions to generate correct weighted bar charts rapidly.

Usage:

from scripts.visuals import plot_weighted_bars

cols = ['Brand_A', 'Brand_B']
plot_weighted_bars(df, cols, "Brand Usage", "chart_brand.png", weight_col='Weight')

Dependencies

Ensure the data analysis stack is installed:

pip install pandas numpy scikit-learn factor_analyzer matplotlib seaborn

Workflow Example (Phase 2 Analysis)

1. •Load Data: Load the csv or sav file.
2. •Weighting: Define universe targets (from project specs) and calculate weights.
3. •Factor Analysis/PCA: Reduce 20+ "Benefit" attributes to 3-4 distinct dimensions.
4. •Segmentation: Cluster users based on these 3-4 dimensions to find "Personas".
5. •TURF: Optimize the feature set for the new product.
6. •Visuals: Export weighted charts for the final PDF report.