Rich-text Reply

Changing total traffic allocation in the middle of a risk-averse experiment, skews results?

lkraav 06-01-15

Changing total traffic allocation in the middle of a risk-averse experiment, skews results?

https://help.optimizely.com/hc/en-us/articles/200040115?flash_digest=d87733fb59bb1559b9e6efcab9061c2... clearly states that messing with individual variation traffic allocation skews results, and that's clear as day. What about moving the total traffic allocation slider around mid-experiment? All variations will get more or less visitors, but what are some result skewing after-effects of this if any?

--
Leho, marketing & tech architect | G+: lkooglizmus@gmail.com
Level 4

Re: Changing total traffic allocation in the middle of a risk-averse experiment, skews results?

Hi Leho,

If you move the total traffic allocation slider around mid-experiment this shouldn't skew the results. You're simply letting less new people into the experiment. All other settings (i.e. percentage of people allocated into specific variations) will remain constant.

Please be aware that it may take you longer to reach statistical significance since velocity of people entering the experiment will slow down. Besides that, you're good to go!

Best,
Amy
Amy Herbertson
Customer Success
JDahlinANF 06-02-15
 

Re: Changing total traffic allocation in the middle of a risk-averse experiment, skews results?

@Amy_Herbertson- We always run our experiments on 100% of traffic, so I haven't looked into this yet, but it seems to apply to this question, so I'll add onto this thread...

 

From this article: https://help.optimizely.com/hc/en-us/articles/200040115

"When you change a variation’s traffic allocation mid-experiment, all new users will be allocated accordingly from then on. However, all users that have entered your experiment prior to the change will be bucketed into the same variation they entered previously, thus altering the results and making it difficult to interpret conversion rate. This is one reason we recommend that people do not alter individual traffic allocation to a single variation."

 

This does not explicitly mention that users who were excluded from the experiment stay excluded.  I presume this to be the case, but wanted to call that out to be sure.