Correlation between survey responses?

Six Sigma – iSixSigma Forums Old Forums General Correlation between survey responses?

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
  • #49345

    CVA BB

    I’m looking for the right tool to show me the relationship betweeen survey responses. The survey asks about statisfaction wtih price, delivery time, and overall experience. We’re trying to see if we can  leverage this existing data to quantify the effect of price and shipping responses on the response to the overall response.
    Responses are on a 1 to 10 scale, and are heavily skewed to the high end (good for us). I can do a simple regression of price and shipping against overall satisfaction, and get plausible results, however, because the data is not continuous, I’m not sure that the inferences from regression are valid. Is there a better tool to quantify the contribution of these inputs to overall satisfaction?



    And what about Mood testing?



    You can use Logistic Regression in this case, to develop your relationships.  Because you are dealing with human’s, the behavior most likely cannot be considered, “normal”, (in the statistical sense!), so your non-parametric analysis tools will be most useful for the analytical side of the project.
    Often, the best tool or, “right” tool, for this type of analysis, will be your graphical tool(s).  Chart it.  Chart it in 2 and 3 axes.  Use your eyes to show you the relationships or lack thereof. 
    And, always run a confirmation test!


    Robert Butler

     As always, rule #1 is plot your data and see what it looks like.  If the pattern is “reasonable”  (and this will be a judgement call) simple regression may be adequate.
     With a 1-10 scale you’re on the ragged edge with respect to the discrete/continuous issue. The post below may help you in your thinking about the value of the above approach.



    Robert:I have sent you a question to your email. Can you check give me a response when you get a chance please…
    Thanks in advance



    The assumptions of linear regression are:
    1. Linear relationship between the IV(s) and DV
    2. Independece of the residuals
    3. Homoscedasticity
    4. Normality of residual distributions
    You can assess (1) linearity by plotting residuals versus predicted values, (2) independence by looking at an autocorrelation plot, (3) homoscedasticity by looking at the residual versus time plot and the residual versus predicted value, and (4) normality of error distributions by looking at a normal probability plot of the residuals.
    Although debatable, most would agree that a variable with a range of 10 can be considered continuous. If your data “reasonably” meet the aforementioned assumptions, then your results should be valid. If they do not meet these assumptions, there are several options (transformations, consider alternative tests).



    I think Robert Butler and Ryan are on target here. If you think of your data as being pixels and colors of a digital image, the number of responses you have from your survey is the number of pixels in your image and your 10-point scale is the number of colors you have to work with. If you have enough pixels, you can make a crude picture that should be recognizable as a human face versus say, a landscape, but you may not be able to identify who’s face it is. That may be acceptable though.
    The point is, you may have enough data to make some intelligent directional decisions, just don’t try to make any subtle inferences.



    As I understand your Y= Overall Satisfaction (1to10-Ordinal data)
    Your x is= Price and Delivery time
    A Simple Regression model will not give correct interpretation.
    I would go for a Oridnal Logistic Regression.
    or Binomial Logistic Regression if I wnat to understand what drives just my top rating (10) only

Viewing 8 posts - 1 through 8 (of 8 total)

The forum ‘General’ is closed to new topics and replies.