Animals are counting on you. There are only 15 days left to join MFA’s Million-Dollar Challenge. Your donation will be matched today. Donate here.

How to Dominate Social Media

Written by Steven Rouk and Nat DeFries

Table of Contents
Short on time? Download the poster with the main results!


We had it beautifully designed to look good taped up beside your desk.

Introduction
Social media is important. How important?

Facebook is used daily by 15% of the world’s population—1.13 billion people. Of all adult internet users in the U.S., 72% use Facebook. Altogether, 70% of Facebook users log on daily, and 43% get on the site multiple times per day. Additionally, 90% of young adults use social media, and 83% of marketers pursue social media initiatives in their strategies (Hubspot.com/marketing-statistics).

These people access social media on a regular basis to consume and create content, and each piece of that content exerts a degree of influence. If we can create influential content that gets consumed by millions of people, we can do an incredible amount of good for animals.

So naturally, Mercy For Animals devotes a lot of attention to social media. We have five Facebook pages (English, Spanish, Portuguese, Hindi-English, and Chinese), three Twitter accounts (English, Spanish, and Hindi-English), and three Instagram accounts (English, Spanish, and Hindi-English). On Facebook, we have 2.5M likes on our English page, 3.1M likes on our Spanish page, 1.5M likes on our Portuguese page, 1.3M likes on our Hindi-English page, and 283K likes on our Chinese page, for a total of around 8.7 million likes on our Facebook pages.

In 2017 alone, we got 270M Facebook video views, 3.5B Facebook impressions, 5.3M shares on Facebook, 4.7M Twitter video views, 140M Twitter impressions, 940K retweets, 2.8M Instagram video views, and 8.8M Instagram impressions.

That’s a lot of people!

With all the people on these platforms and all the time and energy poured into creating content and curating audiences, we have to ask ourselves: Out of all the things we do and post, which of them actually work? If we want to create a high-performing post on one of these platforms, is there a good formula for doing so?

We conducted a study in June 2015 attempting to answer these questions, the results of which can be seen here: MercyForAnimals.org/dominate-facebook. The present study was conducted to assess the robustness of our previous analysis; contribute to research in other countries where animal advocacy is rapidly growing; and determine whether the strategy across Facebook, Twitter, and Instagram should be similar.

In this study we analyzed data from nearly all our social media platforms: all five Facebook pages listed above, as well as the English Twitter and Instagram accounts. The goal was to determine which characteristics were correlated with more engaging posts, as measured by typical social media metrics, such as likes, impressions, and engagement. (See Notes section below for why these specific metrics were chosen.)

We compiled datasets for each platform and language of all our posts from 2016 and then took random samples from these datasets to analyze.

Here are the results.



Key Findings

First, we’ll survey the main results of the most successful posts by social media platform.

And remember—when posting content on social media, always comply with applicable intellectual property laws and the terms of service for the platform.

(See the Notes section below for definitions, clarifications, and other notes.)

Facebook
  1. Original videos performed better than shared videos, photos, links, and plain text in all metrics. Post videos.
  2. Posts about the negatives of animal products and posts about animals performed better in all metrics than posts about health, MFA (organizational) news, other news, and veg foods, as well as humorous posts.
  3. Posts about celebrities and the environment also performed well on some languages’ pages, compared to the same topics listed above.
Instagram
  1. Photos on Instagram resulted in more likes and engagement, but videos provoked more comments.
  2. Posts with short text performed best (under 255 characters).
  3. Posts about dogs were effective.
Twitter
  1. Shared videos from other users or photos were effective.
  2. Posts about the environment, celebrities, or animal advocacy news did well.
  3. Longer tweets did better, up to 166 characters.
  4. Posting about cows worked, but not cruelty footage.
Next, we’ll look at some similarities and differences between the five languages’ Facebook pages.

English Facebook
  1. The English posts received the most impressions, followed by those on the Spanish, Portuguese, Hindi-English, and Chinese pages.
  2. As with all our Facebook pages, videos performed the best by far.
  3. The top-performing content types were posts with celebrities, the negatives of animal-based foods, and animals.
  4. The lowest-performing content type was organizational news, such as news about MFA.
  5. Cruelty footage performed better, as did representations of cows.
  6. Posting in the morning Pacific time seemed to work better.
Spanish Facebook
  1. There was no significant trend related to the hour posted, unlike the English page.
  2. Posting a video worked best.
  3. Including a nonplatform call to action (such as “go vegan”) worked better.
  4. Including representations of cows, but not goats, worked better.
  5. Posting about organizational or animal advocacy news performed worse.
Portuguese Facebook
  1. Videos performed best, as the other pages.
  2. Posting about organizational and animal advocacy news didn’t have a negative effect, as it did on the English and Spanish pages.
  3. Posting about animals worked best, for content. Posts about the negatives of animal foods, the environment, and health had less-conclusive results.
  4. Including a representation of a pig significantly increased impressions more than any single animal in either English or Spanish page data. Posting about dogs and chickens also had some positive effects.
Hindi-English Facebook
  1. Videos performed best in increasing shares and nonfollower impressions, but videos didn’t increase organic impressions.
  2. Posts about animals resulted in one of the largest effects, and posts about celebrities also performed very well.
  3. Humorous posts received fewer shares.
  4. Including cows and goats in the posts didn’t affect organic impressions but did boost shares and nonfollower impressions.
  5. Including a quote decreased impressions and shares.
  6. The more time passed since the previous post, the fewer shares and impressions a post received. (Up to a day between posts.)
Chinese Facebook
  1. Like all other Facebook pages, videos performed the best compared to other content types.
  2. Longer messages received more shares and impressions, even up to 398 characters, which could be around 2,400 characters in English (because of the Chinese logographic writing system).
  3. There were very few significant results related to content type or time posted.
  4. Posting about pigs or cows had a positive effect.
Notes

Interpreting Results
This study looks at social media metrics and doesn’t address outcomes like diet change, attitude change, and willingness to take a certain action. More research is needed on the impact of social media efforts on these kinds of outcomes. We can assume a positive impact on people consuming this media based on related studies about online advertising (see DavidReiley.com/papers/OnlineAdsOfflineSales.pdf), although research specifically addressing the impact of animal advocacy advertising was inconclusive.

Also, the impact of social media posts is likely different from the impact of social media advertising, and in this study we analyzed data from the former. We at MFA would love to see more research into the impact of social media on the outcomes the animal advocacy movement cares about most.

Videos vs. Shared Videos
”Video” always refers to a video that the MFA page posted. This is distinct from “shared video,” which is a video posted by another page and then shared by the MFA page. This analysis will always use the word “share” to indicate shared video content.

Performance and Discussing What Worked
When a category “performed the best” or “worked better,” it resulted in higher totals in the majority of metrics when compared to other categories. While we use more specific language in the analyses below, the key findings are purposely written to be higher-level takeaways that we would suggest for content creators, and thus we use less specific language.

Definition of Facebook Impressions Metrics
Facebook organic impressions: the number of times content was on screen from an unpaid post.
Facebook nonfollower impressions: the number of times content was on screen for someone who is not a follower of the MFA Facebook page.

(Note: These metrics were used based on the view that they would be the most relevant to a truly successful social media post. Anyone can pay to get impressions on their posts, so paid impressions don’t necessarily correlate with good content that resonates with people.)

“Platform Call to Action” vs. “Nonplatform Call to Action”
Nonplatform calls to action might include asking viewers for donations, asking viewers to join a program, or asking viewers to pass out leaflets, while a platform-specific call to action includes any action a user is asked to take using platform functionality, such as liking, sharing, or commenting on a post.

Selection of Social Media Metrics
As briefly noted above, we chose to look at these specific social media metrics because of their usefulness in determining which posts resonated more with the audience. We chose metrics generally used and accepted by social media marketers, while avoiding metrics that correlated too closely with one another. (But we still accidentally picked two metrics that were extremely correlated with each other: engagement and likes for the English Instagram data.)

Use of “Impressions” in Longer Analyses
When "impressions" is used on its own in the in-depth analyses below, the term refers generically to all response variables of interest: organic impressions, nonfollower impressions, and shares.

“Length of Message” Variable in the Chinese Dataset
Chinese uses a logographic writing system in which each logograph represents a word or phrase. This is in contrast to English, Spanish, Portuguese, and Hindi, all of which string together letters from phonetic alphabets to form words. Therefore, Chinese posts have a lot more content than do posts of the same length in other languages.

To convert Chinese message length to the message length in any of the other languages, one can multiply the number of Chinese logographs by the average word length in another language. For example, in English, the average word length is around five characters, six including spaces, which aren't used in Chinese. So a three-logograph message is equivalent to an 18-character message in English; a 398-logograph message, a 2,400-character message.

Sentiment
Where used in the analysis, sentiment was calculated using the Google Natural Language API sentiment-analysis tool.

In-Depth Analyses per Platform and Language
A quick reminder: When posting content on social media, always comply with applicable intellectual property laws and the terms of service for the platform.

English—Facebook
As a follow-up to our 2015 social media analysis, we analyzed data from about 600 Facebook posts randomly sampled from all MFA English Facebook posts in 2016. Facebook is widely used by animal advocacy groups, so it's important to understand how to maximize impressions and reach as many people as possible.

For this and all other analyses, the significance threshold was set at p = 0.05 (95% confidence level).

Analysis Results

Strongest Effects
Type of media and content of post had the strongest effects on impressions.
Posting a video increased organic impressions by 330% and nonfollower impressions by 1,010% compared to posting a link. This trend agrees with the 2015 analysis (MercyForAnimals.org/dominate-facebook). This effect is likely due in part to Facebook's post prioritization algorithm that favors videos over other types of media.

Sharing a video was less effective than posting an original or new one, increasing organic impressions by 150% and nonfollower impressions by 190%. And shared videos got very few (subsequent) shares, 99.7% fewer than posting a link. Photos increased shares by 42% and nonfollower impressions by 34% compared to posting a link but did not significantly affect organic impressions.

Of the many possible content tags, only a few significantly increased impressions. Posts about animal-based foods increased organic impressions by 54% and nonfollower impressions by 125% compared to posts about miscellaneous topics. Posts about animals increased organic impressions by 36% and nonfollower impressions by 99%. Posts featuring celebrities increased organic impressions by 65% and nonfollower impressions by 131%.

Posts about MFA, including MFA news, organizational updates, donation programs, and corporate collaborations, performed the worst. This topic decreased organic impressions by 68% and nonfollower impressions by 66% compared to miscellaneous topics. Humorous and environmental content may also affect impressions; their effects were significant for some response variables.

Time
The day of the week had no significant impact on impressions. Likewise, there was no significant difference between impressions on weekdays versus weekend days.

But time of day had an effect, with highest impressions in the morning (4 a.m. to 10 a.m. Pacific time). Compared to baseline, peak organic impressions were 25% higher and nonfollower impressions were 53% higher. The wide time frame in which impressions were elevated may be due to the different time zones in the U.S.

Unexpectedly, time of year also influenced results. Impressions were highest in spring (February to June). Compared to baseline, peak organic impressions were 36% higher and nonfollower impressions were 44% higher. Time of year influenced all three response variables, so the effect is not likely a statistical aberration. The reason for this trend is unclear, but it may have something to do with changes in Facebook's post prioritization algorithm at that time.

The length of time between posts did not affect impressions.

Animals
Including cruelty footage in a post increased organic impressions by 80% and nonfollower impressions by 158%. Representations of young animals had no apparent effect on impressions, in contrast to the 2015 analysis. Including a representation of a cow increased organic impressions by 29% and nonfollower impressions by 59%, whereas including a chicken decreased organic impressions by 28% and nonfollower impressions by 39%. No other specific animals, including humans, significantly affected impressions.

Veg Foods
There was no difference in impressions between including or not including a recipe in a post or linked media. But this result is inconclusive because the recipe flag variable was inconsistently coded.

Other Effects
Other variables affected impressions but to a lesser extent.

Including a nonplatform call to action decreased organic impressions by 69% and nonfollower impressions by 62%. This is consistent with the 2015 analysis.

The difference between including and not including a quote was not significant. This differs from the 2015 analysis, which found that including a quote increased impressions.

Length of a post’s text had no effect on impressions, in contrast to the 2015 analysis.

Sentiment of a post’s text had no effect on impressions, in contrast to the 2015 analysis. But the 2015 analysis used humans, rather than the Google Natural Language API sentiment-analysis tool used in this study, to determine the sentiment of each post (angry, happy, sad, etc.) and did not include any information about the strength of the emotion.

Summary
You can make a high-performing English Facebook post by the following:
  1. Posting a video
  2. Posting about celebrity support of animal advocacy, veg*ism, etc.; the negatives of animal foods; cute animals; or factory farm cruelty (content tags: celebrity, animal foods, and animals)
  3. Including cruelty footage
  4. Including representations of cows
  5. Posting in the morning Pacific time
Spanish—Facebook
We analyzed data from about 1,300 Facebook posts randomly sampled from all MFA Spanish Facebook posts in 2016.

Analysis Results

Strongest Effects
Type of media, content of post, and inclusion of a nonplatform call to action had the strongest effects on impressions.

As with the English data, type of media included in a post had by far the strongest effect on impressions. Including a video increased organic impressions by 521% and nonfollower impressions by 3,070% compared to including a link. Including a photo increased organic impressions by 15% and nonfollower impressions by 129%. Text-only posts performed the worst of any type (-74% for shares and -68% for nonfollower impressions), which would suggest that it’s best to always include some form of media. Sharing a video, as opposed to posting a new one, may or may not affect impressions. For this dataset, including a shared video had a positive effect on organic impressions, a negative effect on shares, and no significant effect on nonfollower impressions.

Including a nonplatform call to action increased organic impressions by 96% and nonfollower impressions by 204%.

Few post topics were significant in this dataset. In fact, only the MFA news topic affected impressions. Posting about advocacy and organizational news decreased impressions compared to posting about miscellaneous topics. Organic impressions decreased by 52% and nonfollower impressions by 72%. The humor category had a negative effect on shares but was not significant for the other response variables. Similarly, animal foods only affected organic impressions. Although the animals category was not significant in the regression, most of the posts with the highest organic impressions were about animals.

Time
The day of the week had no significant impact on impressions. Likewise, there was no apparent difference between impressions on weekdays versus weekend days.

Unlike the English dataset, the Spanish data showed no significant trend with time of day.

As with the English results, time of year was influential, though less than for the English Facebook data. Impressions were highest in late winter and spring (January through May). Peak organic impressions were 8% higher compared to baseline and nonfollower impressions were 17% higher. Time of year affected all three response variables, so the effect is not likely a statistical aberration. The cause of this trend is unknown, but it may have something to do with changes in Facebook's post prioritization algorithm.

The length of time between posts did not affect impressions.

Animals
Including a representation of a cow increased shares by 24% and nonfollower impressions by 24% but did not significantly affect organic impressions. Including a goat decreased organic impressions by 17% and nonfollower impressions by 29%. Including a cat increased only organic impressions. No other specific animals notably affected impressions. Including either cruelty footage or young animals had no significant effect.

Veg Foods
There was no difference in impressions between including or not including a recipe in a post or linked media. But this result is inconclusive because the recipe flag variable was inconsistently coded, as in the English data.

Other Effects
Including a quote was significant only for shares, which were positively affected.

Including a platform-specific call to action had no significant impact.

Neither number of followers nor length of message had an effect on impressions.

Differences in post sentiment were significant only for shares. A strongly negative post (those that had a sentiment of -1 on a scale of -1 to 1, where 0 is neutral in tone) had 18% more shares than a neutral post. A strongly positive post performed the worst.

Summary
You can make a high-performing Spanish Facebook post by the following:
  1. Posting a video
  2. Including a nonplatform call to action
  3. Not posting about organizational or animal advocacy news
  4. Including representations of cows (and not goats)
Portuguese—Facebook
We analyzed data from about 250 Facebook posts randomly sampled from all MFA Portuguese posts in 2016.

Analysis Results

Strongest Effects
Type of media included in a post and topic of post had the strongest effects on impressions.

As with the English and Spanish data, posting a video garnered the most impressions of any type of media. Organic impressions increased by 350% and nonfollower impressions by 1,370% compared to posting a link. Data from other types were less conclusive. Posting a photo decreased organic impressions but moderately increased nonfollower impressions. Posting a shared video decreased subsequent shares and showed no significant effect on the other response variables.

There were some content effects, but they were different for the Portuguese page than for the English and Spanish pages. Unlike for the English and Spanish pages, posting about animal advocacy and organizational (MFA) news did not decrease impressions. Posting about animals did increase impressions: 49% for organic impressions and 123% for nonfollower impressions. In fact, many of the posts with the highest organic impressions in the dataset featured animals. The effects of other topics were less conclusive. Discussing the negatives of animal foods increased only nonfollower impressions. Discussing the environment increased shares by 218% and nonfollower impressions by 193%. Discussing health increased only shares.

Time
The day of the week had no significant impact on impressions. Likewise, there was no apparent difference between impressions on weekdays versus weekend days.

Time of day was significant for only one response variable.

There was a seasonal effect, though smaller than for the English and Spanish Facebook data. Impressions were highest in winter (November through January). Peak organic impressions were 1.6% higher compared to baseline. Time of year affected only impressions, not shares, so its influence is less clear than for the English and Spanish pages.

The length of time between posts did not affect impressions.

Animals
Including a representation of a pig increased organic impressions by 73% and nonfollower impressions by 172%. This is a much stronger effect of any single animal on impressions than in either English or Spanish data. Including a chicken increased organic impressions but was not significant for the other response variables. Likewise, including a representation of a dog increased shares by 780% but was not significant for the other response variables. No other specific animals significantly affected impressions.

Veg Foods
Posts about vegan and vegetarian foods were not analyzed separately, because of inconsistencies in the coding of food-specific variables.

Other Effects
Number of followers increased organic impressions but was not significant for the other response variables.

Summary
You can make a high-performing Portuguese Facebook post by the following:
  1. Posting a video
  2. Posting about animals and possibly the environment and health
  3. Including a representation of a pig and possibly a dog or a chicken
The results of this analysis suggest that certain issues resonate differently with Portuguese speakers than with English and Spanish speakers. The most striking difference was in the popularity of certain post topics and animal species. For example, posts that included representations of cows had higher impressions with both English and Spanish speakers, while Portuguese speakers preferred pigs. Portuguese speakers also didn't mind posts about animal advocacy and organizational news, which were very unpopular in the other two languages.

Hindi-English—Facebook
We analyzed data from about 250 Facebook posts randomly sampled from all MFA Hindi-English posts in 2016.

Analysis Results

Strongest Effects
Type of media included in a post and topic of post had the strongest effects on impressions.

As with the English, Portuguese, and Spanish data, posting a video garnered the most shares and nonfollower impressions of any type of media, although Hindi-English videos didn’t increase the number of organic impressions. Shares increased by 630% and nonfollower impressions increased by 360% compared to posting a link. The effects of other types of posts were less conclusive. Posting a photo increased shares by 210% and moderately increased nonfollower impressions.

The type of content had a large effect. Posts with animals had one of the largest effects and resulted in more impressions and shares across the board: 73% more organic impressions, 190% more shares, and 150% more nonfollower impressions when compared to the “other” content topic. Similarly, posts with celebrities resulted in 110% more organic impressions, 133% more shares, and 165% more nonfollower impressions. Posts tagged as “other news” (shared news not involving MFA) had 260% more shares and 200% more nonfollower impressions. Content involving the negative effects of animal foods resulted in 180% more shares, and content involving health had 170% more shares, but these content types didn’t receive more impressions. Humorous posts got 67% fewer shares.

Animals
For posts including animals, specific types of animals didn’t appear to have an effect on organic impressions. But including cows and goats did appear to affect shares and nonfollower impressions.

Including cows in a post resulted in 85% more shares and 115% more nonfollower impressions. Including goats resulted in 72% more shares, and there was weak evidence that it also resulted in 68% more nonfollower impressions.

Quotes
Including a quote resulted in a drop in impressions and shares: 48% fewer nonfollower impressions, 43% fewer organic impressions, and 34% fewer shares.

Number of Followers
Counterintuitively, the number of followers the page had correlated with fewer impressions and shares. Our graphs to verify this show that as the number of followers grew steadily over time, impressions and shares gradually decreased.

Time Between Posts
On the Hindi-English page, time between posts had a greater effect than on the other Facebook pages. The longer the period between posts, the fewer impressions and shares a post got, up to a 40% reduction across all measures if a full day passed between posts.

Summary
You can make a high-performing Hindi-English post by the following:
  1. Posting a video
  2. Posting about animals, celebrities, or "other news"
  3. Not including a quote
  4. Posting more frequently
Chinese—Facebook
We analyzed data from about 250 Facebook posts randomly sampled from all MFA Chinese Facebook posts in 2016.

Analysis Results

Strongest Effects
Type of media and length of message had the strongest effects on impressions.

As with the English data, type of media included in a post had by far the strongest effect on impressions. Including a video increased organic impressions by 311% and nonfollower impressions by 1,780% compared to including a link. Including a photo had no effect on impressions. No other types of media were posted.

Length of message was positively correlated with all response variables when messages were between three and 398 characters. A message 398 characters long got 139% more organic impressions and 228% more nonfollower impressions than a message with only three characters.

(See the Notes section for a discussion of message length in the Chinese dataset.)

Time
The day of the week had no significant impact on impressions. Likewise, there was no apparent difference between impressions on weekdays versus weekend days.

Unlike the English dataset, the Chinese data showed no significant trend with time of day.

Again, there was a seasonal effect, though it was smaller than in the English Facebook data. Impressions were highest in spring (March through May, with data not available for January or February). Peak organic impressions were 0.7% higher compared to baseline and nonfollower impressions were 20% higher. Time of year affected all three response variables, so the effect is not likely a statistical aberration. The cause of this trend is unclear, but it may have something to do with changes in Facebook's post prioritization algorithm.

The length of time between posts did not affect impressions.

Animals
Including a representation of a pig increased shares by 68% and nonfollower impressions by 89%; the effect was not significant for organic impressions. Including a representation of a cow increased shares by 88% and nonfollower impressions by 103%; the effect was not significant for organic impressions. Including a representation of a human increased organic impressions by 28% and nonfollower impressions by 54%; the effect was not significant for shares. Of the study’s five Facebook pages, the Chinese page is the only one in which including humans notably affected impressions. Including a young animal increased both shares and nonfollower impressions by 43%. Including cruelty footage had no significant effect.

Veg Foods
This analysis was not performed.

Other Effects
No post topics were significant. This dataset had a larger proportion of posts about animals compared to other topics, making it difficult to draw statistically significant conclusions about the effect of other topics on impressions.

Including a nonplatform call to action decreased organic impressions only; it had no effect on the other response variables.

Number of followers was positively correlated with shares only; it had no effect on the other response variables.

Neither inclusion of a quote nor inclusion of a platform-specific call to action had an effect on impressions.

Sentiment of post caused significant differences in nonfollower impressions. Posts with a strongly negative sentiment (sentiment of -1) had 42% more impressions than tone-neutral posts. Strongly positive posts performed worse than both.

Summary
You can make a high-performing Chinese Facebook post by the following:
  1. Posting a video
  2. Posting a longer message
  3. Including representations of young animals, pigs, cows, or humans
English—Twitter
We analyzed data from about 250 Twitter posts randomly sampled from all MFA English Twitter posts in 2016.

For this and all other analyses, the significance threshold was set at p = 0.05 (95% confidence level). In this analysis, we measured impressions, likes, and retweets.

Analysis Results

Strongest Effects
Type of media included in a post, topic of post, length of post text, and type of animal all had significant effects on the outcomes we measured.

Type
Videos resulted in more impressions, likes, and retweets than links and text, but unlike the Facebook pages, videos weren’t the strongest-performing type. Shared videos performed best on Twitter, resulting in 108% more impressions, 225% more likes, and 165% more retweets compared to link posts.

Photos performed second best, with 83% more impressions, 232% more likes, and 124% more retweets than link posts. And third best, videos resulted in 69% more impressions, 55% more likes, and 124% more retweets compared to a link post.

Content
Posts about the environment performed the strongest on impressions and retweets, with 116% more impressions, 103% more likes, and 222% more retweets than the “other” topic type. But both celebrity posts and “other news” (shared news not involving MFA) posts resulted in more likes than posts about the environment, with celebrity posts getting 85% more impressions, 160% more likes, and 145% more retweets and “other news” posts getting 56% more impressions, 144% more likes, and 150% more retweets.

Performing less strongly, posts about animals resulted in 45% more impressions, 78% more likes, 131% more retweets, and posts about the negatives of animal foods resulted in 62% more impressions and 132% more retweets.

Length of Text
Longer text in tweets correlated with more impressions and retweets. A tweet that was 166 characters long received 35% more impressions and 56% more retweets than a tweet that was 35 characters long.

Animals
Including a representation of a cow resulted in 46% more impressions, 40% more likes, and 67% more retweets. But including cruelty footage resulted in 39% fewer likes.

Summary
You can make a high-performing English post on Twitter by the following:
  1. Posting a shared video or a photo
  2. Making the content about the environment, celebrities, or “other news”
  3. Making the tweet text longer
  4. Including a representation of a cow
  5. Not including cruelty footage
English—Instagram
We analyzed data from about 250 Instagram posts randomly sampled from all MFA English Instagram posts in 2016.

In this analysis we considered engagement and likes.

Analysis Results

Strongest Effects
Type of media in a post and length of post had the strongest effects on engagement and likes.

Unlike the Facebook data, video posts performed worse than photo posts on both engagement and likes: 29% less engagement and 32% fewer likes.

A brief follow-up analysis shows that videos resulted in more comments, but this analysis wasn’t conducted in full and wasn’t included in the original study plan. This would be a good area for future analysis.

Posts with longer text also did worse on both engagement and likes, ranging from 22% less engagement and fewer likes for posts with 255 characters up to 64% less engagement and fewer likes for posts with 1,020 characters.

Animals
When a post included animals, the specific type of animal didn’t matter except when the animal was a dog. Posts with dogs received 40% more engagement than posts with other animals, and weak evidence suggests they also received 38% more likes.

Summary
Overall, there were fewer results for English Instagram posts than for other platforms. From what we learned, though, it appears you can create a high-performing English Instagram post by the following:
  1. Posting a photo (not a video)
  2. Keeping the text in the post short (shorter than 255 characters is best)
  3. Including a representation of a dog
Conclusion
The primary takeaway from this study is much the same as from our 2015 study: Facebook dominates social media and videos dominate Facebook. If you don’t have a strong Facebook presence, you should create one; and if you aren’t creating and posting videos on Facebook, you should start.

But we discovered that videos didn’t perform uniformly strongly across all platforms: Photos on Instagram performed better than videos, and shared videos and photos both performed better on Twitter (although videos still performed much better than text, links, and other types).

Other conclusions were more dependent on the platform and the language. Unsurprisingly, posts about animals did well across the board, but the type of animal with the highest performance changed according to both language and platform.

Analysis Details
We believe in the importance of preregistering studies to help prevent researcher bias in the analysis. Here is a link to our Open Science Framework preregistration.

Our analysis used a series of R scripts to run multiple regressions on social media impressions data. We took the log of the outcome variable to use in the analysis, since the data was highly skewed.

The Boolean variables were coded with a 0/1 flag indicating whether the condition was present, such as “FB CTA” and “Young Animal.”

There were three different outcomes of interest for Facebook (organic impressions, nonfollower impressions, and shares), three outcomes for Twitter (impressions, likes, and retweets), and two outcomes for Instagram (engagement and likes). Preliminary analysis on Facebook was done using organic impressions as the primary outcome.

Analysis 1: All Posts
For this analysis we looked at the relationship between the independent and dependent variables for all posts in the randomly selected samples.

Independent Variables:
  • Post type
  • FB CTA (a call to action specific to the social media platform)
  • Other CTA
  • Quotation
  • Content of post
  • Amount of text
  • Hour posted
  • Day posted
  • Month posted
  • Time between posts
  • Number of followers
Dependent Variables:
Facebook
  • Shares
  • Organic impressions
  • Nonfollower impressions
Twitter
  • Impressions
  • Likes
  • Retweets
Instagram
  • Engagement
  • Likes
Analysis 2: Animal-Only Posts
For this analysis we looked at the relationship between the independent and dependent variables for all animal posts in the randomly selected samples, excluding any posts that were not tagged with the post content type of animals.

Independent Variables:
  • Post type
  • FB CTA (a call to action specific to the social media platform)
  • Other CTA
  • Quotation
  • Cruelty footage
  • Young animal
  • Pig
  • Cow
  • Chicken
  • Goat
  • Fish
  • Cat
  • Dog
  • Wild animals
  • Other (rabbit, donkey, sheep, duck, etc.)
  • Amount of text
  • Hour posted
  • Day posted
  • Month posted
  • Time between posts
  • Number of followers
Dependent Variables:
Facebook
  • Shares
  • Organic impressions
  • Nonfollower impressions
Twitter
  • Impressions
  • Likes
  • Retweets
Instagram
  • Engagement
  • Likes
Analysis 3: Food-Only Posts
This analysis was not performed due to inconsistent coding of the food posts.

Data Processing
The data was processed to remove invalid rows and convert variables to useable classes. Some variables were transformed. For example, the natural logarithm of each response variable was taken to reduce skewness. These ended up as normally distributed variables. The day of the week was converted to an integer, and hour and month were scaled and converted to sinusoids.

For a given dataset, the script ran a linear regression on all valid observations (that is, observations with no missing values for the predictor variables) using a predetermined set of predictor variables and response variables.

The program also automatically generated new fake data to run the model on. The fake data was composed of a series of observations, each of which had all predictors set to reference values, those incorporated into the intercept coefficient. Each observation had a single variable set to a unique nonintercept value such that the fake dataset iterated through a series of variable values for each predictor to represent the variation in the original dataset. The program generated predictions for each response variable on each fake observation. These predictions were used to calculate fraction changes from the reference for that variable. The results of these calculations were marked as significant or not by comparing the p-value for the relevant variable's model coefficient.

The process above was repeated for all response variables using only those observations with post content type of animals.

Acknowledgements and Contact Information
This project involved significant efforts by several people, whom I want to take space to thank.

Huge thanks first and foremost to Animal Charity Evaluators for the grant to complete this project and to Nat DeFries for creating the R scripts and conducting the majority of the analysis.

Thanks also to Krystal Caldwell for guiding the first phase of the project, and to Uno Vaaland for help with coding two datasets’ worth of posts.

Finally, two more big acknowledgements to Samantha Berscht at Animal Charity Evaluators for reviewing this write-up and to our phenomenal editing team at MFA for making the language shine.

Steven Rouk, data analyst at Mercy For Animals. StevenR@MercyForAnimals.org
Nat DeFries, data analyst. NMDefries@gmail.com

Resources
...

If you want to hear about our research blog posts as they're published, see here for instructions on how to do so: How to Subscribe to the MFA Research Blog.

To read more of our research blogs, click here: MFA Research Blog.