It does call into question the representativity of the findings, though. For example, people who always show external images are overrepresented; the influence

It does call into question the representativity of the findings, though. For example, people who always show external images are overrepresented; the influence of the new design on open rates of other users is not predicted reliably. Or, from another angle, I am more likely to load external images in emails that are important to me. The influence of the new design on the open rate of emails that are important to me (which get counted) is probably different from the influence of the new design on the open rate of emails that are NOT important to me (and that don't get counted even if I open them).