You are viewing a single comment's thread from:

RE: Steem: An attempt to show improved distribution to content authors over the past year

in #distribution • 5 years ago (edited)

You're looking at 4-5 data points across an entire year? Or does your chart actually show monthly/weekly/daily fluctuations?

Edit: I apologize, you looked at 1 week ranges. That helps a little with the fluctuations of days, but again, cherry picking weeks like this is not a valid form of trending analysis...

You've cherry picked 4-5 dates that may be the same weekday or may not be the same weekday. If you're comparing Monday vs Saturday vs Wednesday, you're going to see somewhat / drastically different results.

You also picked USA's Halloween, which is a terrible choice. Postings on that date historically are lower than other days that fall on the same day of week.

I appreciate the effort, however, you're not providing enough clarity and consistency.. the only valid dates you can actually consider for an analysis like this are Jan 31 and Jan 31, however, as they're likely on different days of the week, you can't rely too heavily on those direct comparisons either ...

5 years ago in #distribution by
$0.00
    Reply 4
    Sort:  
  • Trending
    • [-]
     5 years ago  

    Thanks for the feedback.

    I wouldn’t call it cherry picking as that suggests I’ve seen the data and chosen the week endings for any other reason than them being 7 days at the end of the month.

    If you're comparing Monday vs Saturday vs Wednesday, you're going to see somewhat / drastically different results.

    There will be one of each of the 7 days In each dataset. Without data to support your argument I’m going to blindly disagree, and prove you wrong later 😛

    How about I expand the results to cover a whole month each time? Would that account for this global Halloween phenomenon you speak of?

    $0.00
      Reply
      [-]
       5 years ago  

      Effectively, my comments were intended to clarify that you are performing a trending analysis on data that should not be considered for such a purpose. It would be like polling Trump supporters as to whether or not they like Donald Trump and extrapolating the results from your survey to say it represents the entire population of voters.

      $0.00
        Reply
        [-]
         5 years ago  

        I expanded the datasets for January 2019 and Jan 2020 in the following post as there was some merit in you suggesting a week in each month was not enough data. The results look very similar.

        $0.00
          Reply
          [-]
           5 years ago  

          I haven't taken a look at that post yet (I apologise) .. I'll take a look

          $0.00
            Reply