Comparing correspondence analysis and the log-ratio alternative for representing categorical data