![]() ![]() Having literature recommendations ranging from. You need to have some understanding as to how good or how bad your loadings are. Some examples are with item selection, Likert scale analyses, and factor loadings.īy the way, I wholeheartedly disagree with the cut off point. But the literature does not have as clear of consensus as you imply. I’m not saying that R’s EFA package is impossible, I’m saying that it is not as intuitive as other packages with similar flexibility, like GAMs or even kernel regression.įor the psychometrics literature, I used Kaiser’s criterion as an example. Exploratory factor analysis is extremely subjective and yet, for even the most fundamental decisions, has no clear theoretical standards.įor software, again I have pretty good programming skills. The truth is I’ve been doing research in statistics for a long time. I used “budding data scientist” as a humility sentence. Good sources are Measurement theory and Applications by Bandalos or, if you want a crash course including the math, Revelle's notes. Focus on material written by experts (i.e. Your problem (which TBH is pretty common) is that you are reading a mix of papers from both experts and applied users. Factor scores are dimensionless variables, so it doesn’t matter if the absolute value changes with the number of items. Don’t use cut offs for factor loadings, but plot them for easier interpretation. Kaiser's criterion is the worst way to select number of factors, we established that like 20 years ago. There is also consensus on many things you have mentioned in psychometrics literature. You need to practice coding more or just use a GUI like Jamovi. Most of what you mentioned suggest a lack of training.Īs for software, this sounds like a lack of programming skills. To be perfectly honest, factor analysis is an advanced technique that you as a "budding data scientist" probably shouldn’t be doing in a first place. ? Bro Likert scales aren’t even continuous how do I expect them to be normally distributed? It has potential but as of right now it seems so disorganized.Įdit: I saw one paper that said in order to use ML procedures your data has to be sufficiently normal. I just don’t get why people like these sets of analyses. I can keep going but I think I’ve proved my point enough. Yet the number of items itself is arbitrary, as shown above! But…if you get rid of it, I may bump your reliability score up just a smidge to get you into your journal’s recommended range.Īlthough it’s comical, it’s still important to realize that one’s factor score is a function of the number of items that factor has. ![]() Turns out I don’t really like variable, although I give little to no justification why. Psss, hey you! Yeah, you! Want a high reliability estimate? Well, you may be in luck. 30.īut reliability analyses just feel so scheme-y. 50 in the social sciences and some saying it can be as low as. I see one paper saying lambda should be higher than. No one seems to agree on what value these loadings should be. I see a version of EFA that uses ordinal data instead, but alas it’s paywalled behind a very expensive, niche statistical software.įactor loadings aren’t spared from criticism. It’s clear that EFA requires continuous scales, but it doesn’t seem to be obvious why using Likert scales is an acceptable replacement. An example is with retaining those factors with eigenvalues greater than one - almost half of the literature accepts this and half rejects it! Yet the most obvious involves Likert scaled items. Not to mention, there are certain techniques that have almost no modern literature support that are considered the gold standard. From gathering items to reliability analysis, the fact that there is a potentially infinite number of models that could work is incredibly scary. R has a function available but, in my opinion, it is incredibly complicated even for me.īut even if you do manage to find a software package, literally every step of the EFA process is so damn subjective. R-bloggers - blog aggregator with statistics articles generally done with R software.Īs a budding data scientist, I love all things statistics except exploratory factor analysis and I can’t wrap my head around why it’s so popular.įor starters, it’s incredibly difficult to find a publicly available software package that can do EFAs with minor headaches.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |