3
$\begingroup$

I'm currently looking at three specific questions of a feedback survey and have been tasked with finding out the characteristics of the lowest scorers, to see if there are any patterns or common characteristics of people who score low on these items. These characteristics are demographic variables (age, sex, education, etc.) as well as other variables specific to the survey.

The way the survey is set up, we use top-box scoring. How top-box scoring works is that each item has a designated optimal answer for participants to choose (Sometimes there are more than one optimal answer if the response is on a scale). An example would be an item that is "Do you feel you were respected throughout the process?". The top-box response to that would be "Yes" since we want participants to feel respected. Therefore, when we talk about low scorers, we are talking about those who did not respond with the top-box response for that specific item.

I use both SAS and R and am thinking of how to explore this, currently in SAS I am using proc freq to look at the scores of these specific items and looking at each variable individually, but I'm wondering if anyone has any ideas for looking at the variables grouped, or any ideas in general.

Thank you!!

Improve this question
$\endgroup$
5
  • $\begingroup$ Can you say more about your data? How many variables (other questions, I guess) are there? Are they all categorical, or are they ordinal / continuous? $\endgroup$
    – gung - Reinstate Monica
    Commented Dec 20, 2023 at 18:50
  • 1
    $\begingroup$ Sure! There are a total of three items to look at. We are looking at each item individually for now, so we will be looking at the first item followed by seeing characteristics of low scorers for that specific item, so on and so forth. All of the items have ordinal responses. Item 1 is "would you recommend this to your family/friends" ranging from definitely no to definitely yes (4 levels), item 2 asks about overall experience ranging from 0-10, and item 3 is a similar question to item 1 with responses ranging from never to always (4 levels) $\endgroup$
    – sixfortyseven
    Commented Dec 20, 2023 at 18:55
  • 1
    $\begingroup$ Because you have a small number of discrete levels (even if you think of them as continuous 'underneath'), making a couple of tables is probably best. I might put the question in rows, and then the other characteristic (eg, sex) in columns, and compute the column-wise proportions (rather than the original counts). At that point, you can skim across the rows and see if the %s seem meaningfully different. You could also use mosaic plots for a quick visual. $\endgroup$
    – gung - Reinstate Monica
    Commented Dec 20, 2023 at 20:41
  • $\begingroup$ I don't have the time to give a detailed answer right now, but a possible option could be multiple correspondence analysis, or possibly factor analysis of mixed data. I don't know about SAS, but there are several R packages implementing these methods. The FactomineR package has a couple of tutorials about it, e.g. factominer.free.fr/factomethods/… $\endgroup$
    – J-J-J
    Commented Dec 21, 2023 at 8:16
  • $\begingroup$ This was great, thank you so much for the info! $\endgroup$
    – sixfortyseven
    Commented Dec 21, 2023 at 19:09

1 Answer 1

Reset to default
3
$\begingroup$

As far as I understand, most of the variables are categorical (nominal or ordinal). There are several ways to explore this data. Here are some options:

  1. As you mentioned, you can get frequency distributions of these variables individually as a first step.
  2. You can check some descriptive statistics. For example, the median and mode might help you to summarize ordinal variables. You can also get summary statistics of variables (e.g., satisfaction) by grouping data based on the characteristics of other variables (e.g., sex).
  3. Since you want to learn about the characteristics of low scorers, I assume you are at least interested in the bivariate association between variables. So, you can create contingency tables. Depending on how you construct these tables (see @gung’s comment), you can get counts as well as row and/or column percentages. You can also create three-way tables but these might be a little harder to interpret.
  4. Visualizations are really helpful in exploratory analysis. Again, as @gung recommended, mosaic plots can be used for two categorical variables. You might want to check related questions on CV, for example this one on two ordinal variables shows different methods.

I am not familiar with SAS but there are lots of options for exploratory data analysis in R. In fact, there are packages solely dedicated to exploratory data analysis. Some of these produce automated reports (e.g., smartEDA) which might be useful for a quick look but I recommend you to create your own tables, visualizations, etc.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.