# Handle data with high variation

I'm analyzing som product data that shows sells over the past day, week and month. In some cases I have a very high number but in most cases this number is quite low, and this is affecting quite a lot my descriptive statistics.

Will this affect my regression analysis? Should I remove data that has some zero (even though not all zeros)?

EDIT: I'm trying to determine which characteristics of the product affect the sales. I have data about some characteristics of the product (e.g., color and category) and then number of sales per product.

• (1) Yes it will, but thid doesn't have to be "bad". (2) Why would you do that? Please edit to give us more detaile (e.g. what is the aim of your analysis). – Tim Oct 30 '17 at 14:19
• @Tim added some additional information – barbara Oct 30 '17 at 14:30
• Regarding your edit: imagine that you have a product in two color versions: red and blue. Imagine that 999 times you sold 0 items of the red product and 1 time you sold 5 items. In case of the blue product, 500 times you sold 1 item and 500 times you sold 2 items. After removind the zeros would you argue that "on average" you sell more items of the red color (5 vs 1.5)? – Tim Oct 30 '17 at 14:32
• @Tim, no that would be incorrect. In my case I have this structure of the data: unique product || sales (monthly) || sales (weekly) || characteristics, so for me it would mean disregard some products that have not been selling/selling badly. Not so sure that it would make it better, just trying to figure out what's the right approach to follow – barbara Oct 30 '17 at 14:42
• I, too, think that we need more information. As a general rule, when most data are small but some very large, it is often worthwhile to think about taking the log() of the data instead of absolute numbers. You'll have to see, whether that suits your situation. – Bernhard Oct 30 '17 at 14:48