0
$\begingroup$

I have two distinct datasets that represent the price of the same product being advertised via two different channels. What I'm trying to determine is if the difference of the sold price on each platform is significant. The datasets would look like this:

Channel1Price   Channel2Price   PriceDifference
        10000          9500            -500
        4000           5000             1000
        5000           5000              0
        8000           7500            -750

I've looked at the different variants of a t-test and f-test. I don't even know if those are the right thing to use for this though. What I want to know is if over the entire dataset the difference between Channel1Price and Channel2Price is significant. The concept I'm having trouble wrapping my head around is that how the difference for each "row" rolls up to reflect the overall dataset.

Improve this question
$\endgroup$
11
  • $\begingroup$ What does each row represent? Is there a natural pairing between 10000 and 9500, or could the Channel2Price column go 5000, 7000, 9500, 7000 (no change to the Channel1Price column) without changing anything? $\endgroup$
    – Dave
    Commented Jun 16, 2020 at 21:41
  • $\begingroup$ Each row represents a distinct product sold for a different day or set of days. You can imagine each row like an airline ticket for a day. Channel1 is Expedia, Channel2 is Kayak. Does that help? $\endgroup$
    – Ty Jones
    Commented Jun 16, 2020 at 21:42
  • $\begingroup$ A different product (e.g. car vs kumquat) or a different instance of the same product (e.g. two kumquats)? $\endgroup$
    – Dave
    Commented Jun 16, 2020 at 22:01
  • $\begingroup$ I think neither actually. The same instance of the same product, just sold on two different platforms. $\endgroup$
    – Ty Jones
    Commented Jun 16, 2020 at 22:14
  • 1
    $\begingroup$ Look into paired t-testing and see if that does about what you want. Report back whether or not it does. $\endgroup$
    – Dave
    Commented Jun 16, 2020 at 23:06

1 Answer 1

Reset to default
0
$\begingroup$

This seems to be a case for the paired t-test. Or, if the price distributions are too far from a normal distribution, use bootstrap or a permutation test.

This as a starting point. There is not very much information in the Q, so conceivably there might be complications we are not told.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.