Others have argued that PLT is a poor metric for web latency.
I was curious: how well do SpeedIndex values correlate with PLT values? The strength of their correlation would give us a sense of exactly how bad PLT is as a metric (assuming that SpeedIndex is what we really care about).
To answer my question I looked at the proportion of PLT to SpeedIndex (for mobile sites):
SELECT
url,
onLoad / SpeedIndex on_load_to_si
FROM
httparchive:runs.2014_05_15_pages_mobile
WHERE
onLoad IS NOT NULL
AND SpeedIndex IS NOT NULL
ORDER BY
on_load_to_si
And the result:
It looks like the correlation is pretty close for most sites. About 15% of the httparchive sites have a better SpeedIndex than PLT relative to the norm, and only a few percent of sites have relatively better PLTs.