Probably not. ELK-style log analysis tools benefit from having messages follow a consistent schema. Using a variable as a key makes indexing much more difficult, and can make it impossible to detect patterns where (for example) a single error appears sporadically across many different values of "thing".

If "thing" were a variable with a small cardinality (like a class name or an enumeration), that might change matters. But I'd still be reluctant to do that; having the two values available in separate fields, rather than as a single key/value pair, is a lot more flexible.