I conducted a survey with 25 5 point questions that resulted in a Likert scale Score variable, while also collecting several ordinal demographic variables: years employment, tasks per year, salary, level of training/education. Each was sorted into different ranges, e.g., 1-2 yrs, 3-5 yrs, 6-10 yrs, 11+ yrs, etc., so even the numerical data (years, dollars) is in an ordinal form with 5-10 levels.
I received around 75 responses. The ordinal variables correlate significantly with each other, Spearman Rank Coefficient varies between 0.3 and 0.6 (no real surprise). When I do a Spearman Rank Coeffecient for each ordinal variable and the Score variable, the values are between 0.25 and 0.4. Two questions:
(1) How could one best calculate and present the independent contribution of each of the demographic variables--e.g., if two persons had the same years of employment, education, and salary, but one did more tasks per year, how would one expect that to impact their Score variable? I fear that my sample size my be too small to say anything definitive, but I would welcome any advice including how this analysis could be done even if it would require a larger sample size.
(2) I am curious if there are any distinct "clusters" of respondents in these variables, but cluster analysis is wholly new to me. I am thinking there might be a handful of outliers, but given the correlations noted I would like to see if distinctive clumps are present in the data. I do wish I could see into 5 dimensions so that I could get a sense of clustering by just looking at the plotted data, but in the absence of that, what approaches to this kind of analysis would you suggest? I know that these data being ordinal rather than numerical significantly reduces the possible response space, but I do think clusters are likely for reasons based around the nature of the survey that are not relevant here.
If relevant: so far this analysis has all just been done in excel for convenience. Years ago I had some training in R but that has almost been completely forgotten, and most of my recent little coding exploits, with little stats work, have been in python.
