Shifting baselines and memory illusions: what should we worry about when inferring trends from resource user interviews?