Stats are not my strong point, but my understanding of ANOVA/ANCOVA is that it's used primarily for experimental design where the IV needs to be a categorical variable, and with a continuous variable as the DV to compare means between different manipulations/conditions.
However, this research article about blended learning effectiveness seems to put the responses to a 5-point likert-scale (eg 'I am satisfied with this [blended course]: strongly disagree ~ strongly agree') as the IV, and 'student grade' as the DV to compare means (pp3-4) using SPSS. The conclusion argued that students who are most satisfied (rated 5) have significantly higher grades than those who are least satisfied (rated 1) (p.4, table 1). I found this puzzling and have a couple of questions..
How is this analysis done in SPSS? Did they put the Likert responses as the IV and then do post-hoc analysis/planned contrast to compare 'strongly disagree' condition to 'strongly agree' condition?
Is it the appropriate analysis to do? Wouldn't a regression to find which questionnaire item best predict grades be a more appropriate tool to use?
I know next to nothing about Structural Equation Modelling, but shouldn't an even more proper analysis technique for this kind of data be SEM? As I understand it, first, they should have done EFA/CFA to confirm that all questionnaire items load into their supposed constructs, then do an SEM to find which construct best predict student grades?
My question seems related to this related question here which suggest that ANOVA is not the right way to go about it? And yet, this particular article got published at a rather well-regarded publication in my field... So, does it indicate a flaw during the peer review process, or am I missing something about the analysis?
(PS: the article is free to view from where I am currently, apologies if the article is paywalled for you. Please suggest what I should do if this is indeed the case)