Spread of species and site scores in R (vegan) DCA ordination

Question

I have performed a Detrended Correspondence Analysis on social survey answers, aiming to see the spread of answers to 25 interview questions and how they related to the three farmer types (n =45) we interviewed. This has been done using 'decorana' in vegan, with the following output:

I'm unsure as to why the sites (which are acting as the sites if this were environmental data, shown as the colored symbols) are on such a different axis than the survey questions (which are acting as the species if this were environmental data), and why they aren't in the middle of the survey questions. Is this an issue, or is this standard for a DCA with so few (45) sites and large variation?

All questions are either set as dummy variables (Married = 0, 1; Single = 0, 1) or are continuous (Age is a range between 18-81 etc).

Any help would be hugely appreciated!!

Answer 1

Both sets of variables have weighted mean of zero. It all depends on weights. Probably OP Income has much higher weight than other variables. DCA is not suited for data that are expressed in different units, such as Married=0, Single=1, Age 18–81 and OP Income in monetary units (dollars?). The sum of each variable is used as its weight (so you cannot either centre your data, or in general, to have negative data). You simply used a method that cannot be used with your data, and you should use some other method instead. PCA with equal scaling of variables is normally used for these kind of data. For DCA all variables must be expressed in same units and be commensurate and non-negative. Still one more time: do not use DCA!

DCA (and CA: Correspondence Analysis) is basically a method that was developed for count data, but it has been stretched for other non-negative data where all variables are measured in same units.