# How to workout statistical significance

My stats has become very rusty so I’m trying to use this study as a way to bring myself back up to speed.

Situation
Analysing website visitors to find out at which groups of people prefer to buy a product (say, Product A) at 2 predetermined prices (say Price X and Price Y).

When a website visitor lands, we already have some data about them, mainly things like what traffic source they came from, device used, keyword, IP address etc.

On our site we know what pages they visited and then ultimately may end up on product page which displays Product A at price X or price Y (we can randomly assign prices or a visitor gets price x, next one get price y and so on). We can also track time spend on each page.

So the data looks like this:
Incoming data -> Home Page -> Product Page A (price x) -> Checkout
Incoming data -> Home Page -> Product Page A (price y) -> Checkout
Incoming data -> Blog Page D -> Home Page -> Product Page A (price x) -> Checkout
Incoming data -> Blog Page D -> Home Page -> Product Page A (price y) -> Checkout
Incoming data -> Blog Page E -> Product Page A (price x) -> Checkout
Incoming data -> Blog Page E -> Product Page A (price y) -> Checkout
Incoming data -> Blog Page D -> Home Page -> Product Page A (price x)
Incoming data -> Blog Page D -> Home Page -> Product Page A (price y)
Incoming data -> Blog Page E -> Product Page A (price x)
Incoming data -> Blog Page E -> Product Page A (price y)

(Of course we have a lot more visitors who do not visit product page and do not proceed to checkout etc).

Questions:
Which groups of people prefer price x?
Which group of people prefer price y?
Which groups are not buyers at price x or price y?
Maybe see this at a stat significance level.

Hard work is not a problem.

• You want A/B testing. Commented Apr 9, 2016 at 5:01
• What I'm confused about is does each path (e.g. Incoming data -> Blog Page E -> Product Page A (price x)) become a variant? So, it's A/B/C... X testing? I'm trying to workout optimal price per cluster of people not a universal optimal price. Hope I'm making sense. Commented Apr 9, 2016 at 5:47
• I'm not looking for an A/B split-test for all people who land on a page... because the incoming data and path might be significant. Commented Apr 9, 2016 at 5:56
• what do you mean by 'group of people' ?
– user83346
Commented Apr 9, 2016 at 7:17
• @fcop clusters of people who have similar attributes. Example: Say people who buy product A at price X mostly come from Facebook and use an MacBook Pro. What I'm trying to do is to find information from all this data and then form hypothesis for further testing. Commented Apr 9, 2016 at 7:20

Speaking from my experience that there might be possibility that purchasing status (success or not) depends upon the

(First Hypothesis:-As people belongs to higher income area have more tendency to more things ) (Second Hypothesis:-If there IP address is very far from your server and you are not using things required to imporve latency and it might be possible that due to High latency the customer close the website. As from my experience if a page is taking more than 5sec than it decrease the company ecommerce sale by 30-40% and some people more likely to buy from the device which is more secured ( Not sure about the percentage of it . But you can consider this if you have data about the customer )

2)Device Used - Here it is quite unclear how much information about the device of customer you are getting. As the mobile device during these days are driving 50% of e-commerce economy

3) Page Movement -Check How customer are moving to your webpage i.e. are they moving from a page where you advertised or they come to you via google search. what percent of those that are moving from different sources end up buying through your website ? Are your conclusion statistically significant ?. By going from this way you can conclude the probability that a particular person ends up buying some stuff from your site

Now considering these things in mind and many more (use your intellectualism) you can segment a person according to buying behaviour .You can use various classification technique to segment your customer like LDA , QDA , KNN which ever suits you best.