ruxt icon indicating copy to clipboard operation
ruxt copied to clipboard

Expose ECT for sites

Open jamesshannon opened this issue 8 years ago • 3 comments

HI,

Not sure if this should be in front-end or back-end (guess it should technically be in both), but one of the things I've used CrUX data for, which I'd like to see this expose, is effective connection type.

  1. What percentage of users are on (effective) 3G connections? I find that most people assume "my users are in the US and Europe -- everybody is on 4G" though that's obviously not true and CrUX data can provide a sense of how untrue that is. I've used this query before:
  SELECT
 origin,
 form_factor.name as form_factor,
 effective_connection_type.name as ect,
 SUM((
   SELECT
     SUM(bin.density)
   FROM
     UNNEST(first_contentful_paint.histogram.bin) bin
   WHERE origin IN ('https://www.bmwusa.com', 'https://www.infinitiusa.com', 'http://www.lincoln.com'))) AS density
FROM
 `google.com:chrome-ux-report-test.chrome_ux_report.20171018_preview_origin`
 group by form_factor, origin, ect
 HAVING density > 0 

jamesshannon avatar Dec 19 '17 16:12 jamesshannon

Thanks for the suggestion!

Your motive is interesting, but I am afraid there is a slight catch. Just because, say, 40% data of example.com is from 3G in CrUX report, can we fairly say that represents the actual %?

Nonetheless, we can always expose which % of data has the selected dimensions (eg. 30% of total data is phone+3G, when phone and 3G are selected as dimensions). The conclusion is up to the user of the tool.

I will put this expected feature in our pipeline.

RatulSaha avatar Dec 19 '17 18:12 RatulSaha

Depends on how you define "data" and "actual %". IIRC, this data is based on requests. So it should accurately represent the "actual % (of requests)" (at least with the caveat of those who have opted-in, and to the same extent that the FCP and load metrics are accurate).

Where it gets tricky is trying to conflate requests with users. You're likely to see a lot less requests per user (while on 3G), so this will undercount # of users and sessions (same as with the speed data that you present).

jamesshannon avatar Dec 19 '17 18:12 jamesshannon

Good point. I am not sure if a data point is equated with requests or sessions. It has been announced that the effective connection type is collected through Network Information API.

Let us circle back at it.

RatulSaha avatar Dec 19 '17 18:12 RatulSaha