AI-Enabled Samsung Galaxy Z Series with Innovative Foldable Form Factor & Significantly Improved Screen Delivers New User Experiences Across Productivity, Communication & Creativity The…
How has the effectiveness of different kinds of Facebook post changed in the past 3.5 years?
Using a database of 11.4 million posts published by just over 24,000 of the most liked Facebook pages, dating back to 2010 and earlier, I was able to analyse the performance (in terms of likes) of the four most common post types: Photo, Video, Link, and Status (text only posts) over three and a half years.
My earlier research showing that Photo posts typically receive the most likes was confirmed by this newer, larger study, but I also found a few other interesting trends. Simple text-based posts have been on a decline since early 2012 and the Video type has begun to gain effectiveness since late 2013.
There was also an interesting shakeup in the order of post type performance in late 2011, right around the time Facebook was rolling out a number of changes, including timeline profiles, the ability to “like” anything, and the introduction of the Ticker. It appears they also experimented with adjusting the weighting of the variables used to calculate how much of each post types their users saw. A change that seems to have been rolled back by mid 2012.
For the geeky details on the actual calculations I did, check out the graph below.
For each page I calculated the average number of likes on posts they published in each of the 43 months from January 2010 to July of 2014. I then compared the number of likes attracted by posts of the four major types to those per-page monthly averages. For each type type I calculated a percentage lift (or drop) above (or below) the average. The percent change numbers were averaged across the dataset for each month and presented in the graph you see below.
The calculations I performed allowed me to control for changes in page audience and overall engagement rate and isolate the relationship between likes and post types. This is the most straightforward way I could accomplish this in the absence of historic total page like data for all of the pages in my dataset.
This article by Dan Zarella originally appeared on Danzarella.com and is republished with permission.