Factors affecting customer satisfaction
0 likes•127 views
The project analyzes customer satisfaction by examining 12 variables through descriptive analysis, regression modeling, and principle component analysis. It identifies three key factors—order management, marketing attributes, and product attributes—that significantly contribute to customer satisfaction, achieving an adjusted R-squared of 86.27% in the final regression model. The findings suggest that the company can focus on these three factors to improve customer satisfaction, minimizing the need for the original 11 variables.
Factors affecting customer satisfaction
- 1. Project – 2 Factors affecting Customer Satisfaction (Principle Component/Factor Analysis & Regression) - By Saleesh Satheeshchandran
- 2. Table of Contents 1.Variable Expansions 2.Project Objective 3.Exploratory Analysis 4.Inspection for Multicollinearity 5.Simple linear regression 6.PCA/Factor analysis 7.Interpreting the output and naming the Factors 8.Multiple linear regression 9.Conclusion 10. Reference Variable Expansions
- 3. Project Objective The objective of this project is to analyze the data of a company where there are 12 variables available. 11 independent variables and one dependent variable. Initially a descriptive analysis needs to be done and later a model needs to be created that would help in predictive analysis. Ultimately, this would help in the improvement of the dependent variable, namely, customer satisfaction. Exploratory Analysis - The data in .csv format was uploaded into the R studio. - The summary shows 12 variables - The data length of each variable is 100 - All the data variables are numeric in nature. - Minimum, Maximum, Mean and the quartile ranges show that all the variables have a value that is captured in a 10-point scale (with single decimal point system). - There are no missing values in the given data - Box plot of the variables show the below status on outlier values. - - ProdQual – No Outliers - - Ecom- 4 outlier values - - TechSup– No Outliers - - CompRes– No Outliers - - Advertising– No Outliers - - ProdLine– No Outliers - - SalesFImage- 2 outlier values - - ComPricing– No Outliers - - WartyClaim– No Outliers - - OrdBilling- 3 outlier values - - DelSpeed- - 1 outlier value - - Satisfaction– No Outliers
- 5. Inspection for Multicollinearity - Correlation table and correlation plot between various independent variables clearly shows that there is a multicollinearity among them. - Warranty claim and Tech support have a .75 correlation - SalesFImage and Ecom has .79 correlation - CompRes and DelSpeed are correlated by .86 - A multi variable linear regression was done and its VIF (Variance inflation factor) calculated. - One of the VIF values is clearly high (DelSpeed),which shows the presence of multicollinearity. - - -
- 7. Simple linear regression 1. Product Quality Linear regression Equation – Y=3.675 + 0.415 * X Where Y is the Satisfaction and X is the Product Quality Very low p value shows the model is significant. Adjusted R squared shows the model explains 22.87% of the variability of the Y value.
- 8. 2. Ecom Linear regression Equation – Y=5.151 + 0.481 * X Where Y is the Satisfaction and X is the Ecom Very low p value shows the model is significant. Adjusted R squared shows that the model explains 7% of the variability of the Y value. 3. TechSup Linear regression Equation – Y=6.447 + 0.0876 * X Where Y is the Satisfaction and X is the TechSup From the p value, the model is significant. Adjusted R squared shows that the model explains 7% of the variability of the Y value.
- 9. 4. CompRes Linear regression Equation – Y=3.680 + 0.595 * X Where Y is the Satisfaction and X is the CompRes Very low p value shows the model is significant. Adjusted R squared shows model explains 35.7% of the variability of the Y value. 5. Advertising Linear regression Equation – Y=5.625 + 0.322 * X Where Y is the Satisfaction and X is the Advertising Very low p value shows the model is significant. Adjusted R squared the model explains 8.3% of the variability of the Y value.
- 10. 6. ProdLine Linear regression Equation – Y=4.022 + 0.498 * X Where Y is the Satisfaction and X is the ProdLine Very low p value shows the model is significant. Adjusted R squared the model explains 29.6% of the variability of the Y value. 7. SalesFImage Linear regression Equation – Y=4.070 + 0.555 * X Where Y is the Satisfaction and X is the SalesFImage Very low p value shows the model is significant. Adjusted R squared the model explains 24.26% of the variability of the Y value.
- 11. 8. ComPricing Linear regression Equation – Y=8.038 – 0.160 * X Where Y is the Satisfaction and X is the ComPricing Very low p value shows the model is significant. Adjusted R squared the model explains 3.3% of the variability of the Y value. 9. WartyClaim Linear regression Equation – Y=5.358 – 0.258 * X Where Y is the Satisfaction and X is the WartyClaim Very low p value shows the model is significant. Adjusted R squared the model explains 2.1% of the variability of the Y value.
- 12. 10. OrdBilling Linear regression Equation – Y=4.054 – 0.670 * X Where Y is the Satisfaction and X is the OrdBilling Very low p value shows the model is significant. Adjusted R squared the model explains 26.5% of the variability of the Y value. 11. DelSpeed Linear regression Equation – Y=3.279 – 0.936 * X Where Y is the Satisfaction and X is the DelSpeed Very low p value shows the model is significant. Adjusted R squared the model explains 32.62% of the variability of the Y value.
- 13. Principle Component/Factor Analysis - Correlation There are 11 independent variables in the given case. We need to crunch down these variables into a limited number of principle components so that a meaningful regression model can be achieved. Also, there is significant multicollinearity between the independent variables. Hence, they cannot be considered truly independent. Thereby, making it necessary to do a factor analysis. - Eigen Values Checking the Eigen value of the correlation of the variables help in ranking the components in the model.
- 14. - PCA Methodology - Scree Plot Scree Plot helps in identifying the Principle components using Elbow rule. As per Kaiser Rule, only 3 factors are sufficient to be taken in this model. Because, these 3 are having value above 1. But since there is clear instruction from the company to go for 4 components, we are taking the same.
- 15. - PCA output There are 4 Principle components. PC1, PC2, PC3 and PC4 - PC1 accounts for 42.53% of the variation in the dependent variable. - PC1 + PC2 accounts for 69.39% - PC1 + PC2 + PC3 accounts for 87.19% - PC1 + PC2 + PC3 + PC4 accounts for 100%
- 16. Factor Analysis Methodology – Orthogonal rotation Interpreting the output and naming the Factors There are 4 Factors resulting from orthogonal rotation. RC1, RC2, RC3, RC4 RC1- Accounts for 33.19% of the variation in the dependent variable. RC1+RC2 – Accounts for 59.17 of the variation RC1+RC2+RC4 – Accounts for 80.33% of the variation RC1+RC2+RC4+RC3 – Accounts for 100% of the variation. 1. RC1 to be named Order.Management Because, the factor has greater loading from Delivery speed, Order Billing and Complaint Resolution. 2. RC2 to be named Marketing.Attributes Because, the factor has greater loading from Sales Force image, E-commerce and Advertising. 3. RC3 to be named Aftersales Because, the factor has greater loading from Tech support, Warranty & Claim. 4. RC4 to be named Product.Attributes Because, the factor has greater loading from Product quality, Product Line and Competitive Pricing.
- 17. Multiple linear regression of the factors Extracting the data against the Factors to create new table for Regression analysis Harman method is employed to obtain the factor scores. These scores are copied and pasted in an Excel sheet. Factor names are used to replace the variable names. The data is formatted and the sheet is saved as .csv file. The . CSV file is uploaded into R Studio for MLR Analysis.
- 18. Descriptive Analysis of the Factor Data. Descriptive analysis is done to check the nature of the data. - The correlation data shows no significant correlation between the independent variables. - The plot shows no significant correlation between the independent variables. - The correlation data shows sufficient correlation between the independent variable and dependent variable for each of the factor. -
- 19. First MLR Analysis - A Multiple linear regression is done on the factor data - VIF values are calculated. - None of the VIF values have high value (Greater than 5) - Summary analysis is done to check significance - One of the variables, ‘Aftersales’ has a high P value and hence is deemed non- significant. - Class of the variable shows it is a numeric data. - The summary of the variable shows a continuous nature. - Since the non-significant variable is continuous in nature, it is deleted from the model.
- 20. Second MLR Analysis After removing the non-significant variable another round of MLR is done Multiple Linear Regression Model - Y=6.918+0.610*X1+0.639*X2+0.777*X3 Where, X1 - Order management, X2 - Marketing Attributes, X3 - Product Attributes All the 3 factors show a very high significance. Adjusted R square shows 86.27% of the dependent variable being answered by the model. All the 3 factors show similar standard errors, t-values and p-values Conclusion - A Multiple Linear Regression model is achieved by using factor data. - In the future customer satisfaction can be predicted using this regression model. - The company can eliminate the use of 11 variables to measure the customer satisfaction. - Instead they can go for only those variables that contribute to the 3 factors which are order management, marketing attributes and Product attributes. - Maximizing the above 3 factors will maximize customer satisfaction. Reference All the reference R-Screens are pasted along with the observations.