Question

Topic: Research/Metrics

Analyzing Customer Satisfaction Data

Posted by Anonymous on 1/2/2008 at 10:27 PM ET 125 Points
Dear Experts,
I have one question in relation to how can we analyze customer satisfaction data , suppose I have 8 to 10 close-ended questions ( extremely satisfied , ….. , extremely unsatisfied) with four demographics questions age , sex , ,,. etc.
What is the right statistical technique that can be used to deal with such data?
Many thanks in advance
Omar
To continue reading this question and the solution, sign up ... it's free!
SUBSCRIBE
Or, SIGN IN

RESPONSES

  • Posted by wnelson on 1/2/2008 at 10:49 PM Member
    Omar,

    One technique is conjoint analysis. Software programs like Systat and others have the capability to input data and conduct the analysis.

    However: The software will give you hints as to interpretation of the results, but unless you are a statistical expert, you may miss something. If you are not a statistical expert, you may seek one to help you with the analysis. The statistical expert should be brought in to help design the questions to remove any biases also.

    I hope this helps.

    Wayde
  • Posted by Candureactor on 1/4/2008 at 4:59 PM Member
    I'd suggest starting by exploring your data.

    First, run a frequency (count) on each of your demographic questions. Are the results what you expected? For example, do the results compare to your customer base or overall population? If not, that suggests that you may have some sampling issues or some respondent bias. You can correct for that, but you may also need to consider redoing the survey with a better sampling strategy. If your customers are mainly men and your respondents are mainly women then you have a problem with the data that requires a redo.

    If not, then calculate the frequency and mean score for each of your closed-ended questions. Rank these questions and see if that tells you anything. For example, satisfaction might be high but pricing is relatively much lower.

    Third, cross tabulate your demographic data with your close-ended questions. Are there some populations with higher or lower scores than others? What does that suggest?

    Once you have these basic “freqs” or frequencies you can start to do some more number crunching. This may require some special software and people with expertise to run these functions. I’d recommend running:

     A correlation between your criterion variable (e.g. overall satisfaction) and all the other close-ended questions;
     The coefficient of correlation to be able to rank order the variables that have the most effect;
     Creating a table that plots mean scores by the coefficient of correlation for the close-ended questions
     A regression analysis of the close-ended variables if you have more than a few hundred cases.

    While the more sophisticated analysis is cool, you can certainly get a lot from the first three suggestions. All the best and have a happy new year.
  • Posted by Candureactor on 1/7/2008 at 9:33 AM Accepted
    Omar:

    1. I like to plot the mean score of a variable on one axis against the correlation of coefficient (correlated against your criterion variable such as satisfaction) on another resulting in a scatter plot for the 8-12 variables. You can then create a matrix by adding a horizontal and a vertical line through the middle of the plot points (its a subjective decision on where to put them). Those attributes that have high mean scores and high correlation of coefficient will appear in the top right box. These are the scores that are driving up satisfaction. The attributes with low mean scores and high correlation of coefficient will appear in the top left box. These are the attributes that are driving down satisfaction.

    2. Linear regression is suitable to this type of analysis. In simple terms, regression, unlike correlation of coefficient, accounts for multi-collinearity. You need to do some experimentation to be able to load the variables in the correct order and eliminate those variables that cloud the model. Like any social science modeling you are trying to work to "best fit" not "perfect fit" because regression does not account for attributes you didn't measure. Also, be careful with regression as it does not explain directionality. For example, you may find that customer satisfaction is related (large beta weight) to product color... actually it could be that color effects satisfaction or that satisfaction effects color. You need to be able to think this through and make your logical assumptions known to your client.

    Good luck.
  • Posted by Candureactor on 1/8/2008 at 10:43 AM Member
    Hi Omar:

    You have again two questions.

    1. Do mean scores for likert questions give an accurate picture of what your customer thinks about your company's products and services? Tough question and the answer is yes and no. Yes, because research is a good tool but not a perfect tool. No because:

    a. Did you talk to the right people and enough people to get a representative sample?
    b. What is the margin of error / confidence interval for your sample? Here is a calculator: https://www.dimensionresearch.com/resources/calculators/conf_means.html
    c. Are there biases that were introduced because of the way the question was phrased, the methodology (how it was asked), or because of social acceptability (e.g. are people embarrassed by the subject), or other cultural biases?
    d. Are there other attributes that you didn't ask about that are more influential on customer opinion? For example, location (drive time, parking, convenience) is very important for retail customers and if you didn't ask you won't know what impact it has on customer satisfaction.

    2. Spearman: yes, this is fine. Put your criterion measure at the top of the list (e.g. satisfaction).

    Your welcome.
BACK TO TOP

Post a Comment

Related Discussions

Community Info

Top 25 Experts
(Research/Metrics)