Posting Frequency and Engagement

Overview

Many of us have heard that posting consistently is one of the most important things you can do to grow your audience on social media, and in this analysis we’ll attempt to ascertain whether that is true and quantify the effect that posting frequency actually has.

We’ll start the analysis by looking at follower growth. What effect does posting frequency have on follower growth?

Conclusions

This analysis of 4.8 million channel-week observations across Facebook, Instagram, and X provides clear evidence that posting frequency directly impacts follower growth. Tthe Z-score analysis and fixed effects regression model tell the same story: posting more leads to better performance.

The data reveals several key insights to keep in mind. First, there’s a penalty for not posting. Channels that don’t post in a given week consistently underperform their baseline growth rates, which suggests that people and algorithms reward active accounts and penalize inactive ones.

Along those lines, any posting is way better than no posting. Even channels that post just 1-2 times per week see meaningful improvements over weeks when they don’t post at all. To me this suggests that consistency matters more than anything. There’s a real cost to not posting anything.

Third, the benefits of posting more appear to grow at higher posting frequencies rather than diminish. Channels posting 10+ times per week see the largest follower growth, adding an average of 32 additional followers compared to weeks with no posts.

The fixed effects model is particularly compelling because it compares each channel against itself over time, controlling for all the factors that make channels different from each other. When the same channel posts more frequently, it consistently grows faster. This gives us confidence that posting frequency is actually driving growth, not just correlated with it.

For creators, this analysis supports a strategy of frequent and consistent posting. While the optimal frequency varies by platform and audience, the data clearly shows that more active channels outperform less active ones. The question isn’t whether to post frequently - it’s how frequently you can post while maintaining quality and sustainability.

Posting Frequency and Follower Growth

Buffer collects followers data for Instagram, Facebook, and X profiles that have the analytics entitlement, meaning they belong to users that have a paid plan or started a trial. We also collect data for LinkedIn pages, but not personal profiles. For that reason, we’ll focus on Facebook, Instagram, and X audience growth.

The SQL query below returns approximately 4.8 million records that include the number of posts each channel creates per week as well as their weekly follower growth. Approximately 161k profiles are included in this dataset.

sql <- "
with profile_activity_windows as (
  -- Find the first and last week of posting activity for each Facebook profile
  select 
    p.service_id as channel_id,
    'facebook' as service,
    min(timestamp_trunc(up.sent_at, week)) as first_post_week,
    max(timestamp_trunc(up.sent_at, week)) as last_post_week
  from dbt_buffer.publish_updates as up
  inner join dbt_buffer.publish_profiles as p
    on up.profile_id = p.id
  where up.profile_service = 'facebook'
    and up.sent_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1, 2
  
  union all
  
  -- Find the first and last week of posting activity for each Twitter profile
  select 
    p.service_id as channel_id,
    'twitter' as service,
    min(timestamp_trunc(up.sent_at, week)) as first_post_week,
    max(timestamp_trunc(up.sent_at, week)) as last_post_week
  from dbt_buffer.publish_updates as up
  inner join dbt_buffer.publish_profiles as p
    on up.profile_id = p.id
  where up.profile_service = 'twitter'
    and up.sent_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1, 2
),

weekly_facebook_posts as (
  select 
    p.service_id as channel_id
    , 'facebook' as service
    , timestamp_trunc(up.sent_at, week) as week
    , count(distinct up.id) as posts
  from dbt_buffer.publish_updates as up
  inner join dbt_buffer.publish_profiles as p
    on up.profile_id = p.id
  where up.profile_service = 'facebook'
    and up.sent_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_instagram_posts as (
  select 
    service_id as channel_id
    , 'instagram' as service
    , timestamp_trunc(created_at, week) as week
    , count(distinct service_update_id) as posts
  from dbt_buffer.analyze_instagram_analytics_user_media_totals
  where created_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_twitter_posts as (
  select 
    p.service_id as channel_id
    , 'twitter' as service
    , timestamp_trunc(up.sent_at, week) as week
    , count(distinct up.id) as posts
  from dbt_buffer.publish_updates as up
  inner join dbt_buffer.publish_profiles as p
    on up.profile_id = p.id
  where up.profile_service = 'twitter'
    and up.sent_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_facebook_followers_base as (
  select 
    service_id as channel_id
    , 'facebook' as service
    , timestamp_trunc(checked_at, week) as week
    , max(fans) as week_end_followers
  from dbt_buffer.analyze_facebook_analytics_page_daily_totals
  where checked_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_facebook_followers as (
  select 
    channel_id
    , service
    , week
    , week_end_followers
    , lag(week_end_followers) over (partition by channel_id order by week) as prev_week_followers
  from weekly_facebook_followers_base
),

weekly_instagram_followers_base as (
  select 
    service_id as channel_id
    , 'instagram' as service
    , timestamp_trunc(checked_at, week) as week
    , max(followers_count) as week_end_followers
  from dbt_buffer.analyze_instagram_analytics_user_daily_totals
  where checked_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_instagram_followers as (
  select 
    channel_id
    , service
    , week
    , week_end_followers
    , lag(week_end_followers) over (partition by channel_id order by week) as prev_week_followers
  from weekly_instagram_followers_base
),

weekly_twitter_followers_base as (
  select 
    service_id as channel_id
    , 'twitter' as service
    , timestamp_trunc(checked_at, week) as week
    , max(followers_count) as week_end_followers
  from dbt_buffer.analyze_twitter_analytics_user_daily_totals
  where checked_at >= timestamp_sub(current_timestamp, interval 365 day)
  group by 1,2,3
),

weekly_twitter_followers as (
  select 
    channel_id
    , service
    , week
    , week_end_followers
    , lag(week_end_followers) over (partition by channel_id order by week) as prev_week_followers
  from weekly_twitter_followers_base
),

combined_data as (
  select 
    f.channel_id
    , f.service
    , f.week
    , coalesce(p.posts, 0) as posts
    , f.week_end_followers
    , f.prev_week_followers
    , case 
        when f.prev_week_followers > 0 
        then ((f.week_end_followers - f.prev_week_followers) * 100.0 / f.prev_week_followers)
        else null 
      end as follower_growth_pct
    , f.week_end_followers - f.prev_week_followers as follower_growth_absolute
  from weekly_facebook_followers f
  left join weekly_facebook_posts p on f.channel_id = p.channel_id and f.week = p.week
  -- Only include weeks within the active posting window
  inner join profile_activity_windows paw 
    on f.channel_id = paw.channel_id 
    and f.service = paw.service
    and f.week >= paw.first_post_week 
    and f.week <= paw.last_post_week
  
  union all
  
  select 
    f.channel_id
    , f.service
    , f.week
    , coalesce(p.posts, 0) as posts
    , f.week_end_followers
    , f.prev_week_followers
    , case 
        when f.prev_week_followers > 0 
        then ((f.week_end_followers - f.prev_week_followers) * 100.0 / f.prev_week_followers)
        else null 
      end as follower_growth_pct
    , f.week_end_followers - f.prev_week_followers as follower_growth_absolute
  from weekly_instagram_followers f
  left join weekly_instagram_posts p on f.channel_id = p.channel_id and f.week = p.week
  -- Note: Instagram uses a different data source, so no activity window filtering here
  -- unless you want to create a similar window based on the instagram posts data
  
  union all
  
  select 
    f.channel_id
    , f.service
    , f.week
    , coalesce(p.posts, 0) as posts
    , f.week_end_followers
    , f.prev_week_followers
    , case 
        when f.prev_week_followers > 0 
        then ((f.week_end_followers - f.prev_week_followers) * 100.0 / f.prev_week_followers)
        else null 
      end as follower_growth_pct
    , f.week_end_followers - f.prev_week_followers as follower_growth_absolute
  from weekly_twitter_followers f
  left join weekly_twitter_posts p on f.channel_id = p.channel_id and f.week = p.week
  -- Only include weeks within the active posting window
  inner join profile_activity_windows paw 
    on f.channel_id = paw.channel_id 
    and f.service = paw.service
    and f.week >= paw.first_post_week 
    and f.week <= paw.last_post_week
)

select 
  service
  , channel_id
  , week
  , posts
  , week_end_followers
  , prev_week_followers
  , follower_growth_pct
  , follower_growth_absolute
  , case 
      when posts = 0 then 'No Posts'
      when posts between 1 and 2 then '1-2 Posts'
      when posts between 3 and 5 then '3-5 Posts'
      when posts between 6 and 10 then '6-10 Posts'
      when posts > 10 then '10+ Posts'
    end as posting_frequency_bin
from combined_data
where prev_week_followers is not null
order by service, channel_id, week
"

# get data from BigQuery
posts <- bq_query(sql = sql)

We can use the skim function from the skimr package to get a summary of this dataset.

# remove NA values
posts <- na.omit(posts) %>% 
  mutate(posting_frequency_bin = factor(posting_frequency_bin, 
                                        levels = c("No Posts",
                                                   "1-2 Posts",
                                                   "3-5 Posts",
                                                   "6-10 Posts",
                                                   "10+ Posts")))

# skim posts dataset
skim(posts)

Data summary

Name	posts
Number of rows	4761210
Number of columns	9
_______________________
Column type frequency:
character	2
factor	1
numeric	5
POSIXct	1
________________________
Group variables	None

Variable type: character

skim_variable	n_missing	complete_rate	min	max	empty	n_unique	whitespace
service	0	1	7	9	0	3	0
channel_id	0	1	4	19	0	158755	0

Variable type: factor

skim_variable	n_missing	complete_rate	ordered	n_unique	top_counts
posting_frequency_bin	0	1	FALSE	5	No : 2539606, 1-2: 1061589, 3-5: 621423, 6-1: 279856

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
posts	0	1	3.68	25.20	0	0	0	2.00	4695	▇▁▁▁▁
week_end_followers	0	1	21708.63	337551.29	-1938	243	1044	4100.00	56666742	▇▁▁▁▁
prev_week_followers	0	1	21681.23	337541.27	1	242	1040	4086.00	56699599	▇▁▁▁▁
follower_growth_pct	0	1	4.39	1263.37	-14500	0	0	0.23	1496600	▇▁▁▁▁
follower_growth_absolute	0	1	27.40	999.91	-1343728	0	0	3.00	418141	▁▁▁▇▁

Variable type: POSIXct

skim_variable	n_missing	complete_rate	min	max	median	n_unique
week	0	1	2024-06-30	2025-06-22	2024-11-24	52

Next we’ll calculate the average follower growth rate by the number of posts sent in a given week and plot them for each network.

The plot a strong positive correlation for each network. It also shows a potential penalty for not posting, particularly for X profiles. This is the sort of correlation we might expect, but we still want to control for the differences in profiles’ natural growth rates.

# summary stats
posts %>% 
  group_by(service, posting_frequency_bin) %>% 
  summarise(avg_growth = mean(follower_growth_pct, na.rm = T)) %>% 
  ggplot(aes(x = posting_frequency_bin, y = avg_growth, fill = service)) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~service, scales = "free_y") +
  labs(x = "Weekly Number of Posts",
       y = NULL,
       title = "Average Follower Growth Rate by Posting Frequency",
       subtitle = "Profiles that post more frequently tend to gain more followers on average")

To control for these difference we’ll first utilize Z-scores. This is the approach to calculate a Z-score, which we’ll do for each channel:

1. Calculate the mean and standard deviation of follower growth rates across all weeks
2. Calculate a Z-score for each week. Z = (follower growth - average growth) ÷ standard deviation.

For each channel and week, the Z-score tells us how many standard deviations above or below average a channel performed in a given week:

Z = 0: Typical performance for this channel Z = +1: Strong positive week (better than ~84% of weeks) Z = +2: Exceptional week (better than ~97% of weeks) Z = -1: Poor week (worse than ~84% of weeks)

Below we calculate z-scores for each profile each week and calculate some summary statistics.

# calculate channel-specific mean and standard deviation
channel_stats <- posts %>%
  group_by(channel_id) %>%
  summarise(
    mean_growth = mean(follower_growth_pct, na.rm = TRUE),
    sd_growth = sd(follower_growth_pct, na.rm = TRUE),
    n_weeks = n()
  ) %>%
  # only keep channels with sufficient data and variation
  filter(n_weeks >= 3, sd_growth > 0)

# merge back and calculate Z-scores
posts_with_z <- posts %>%
  inner_join(channel_stats, by = "channel_id") %>%
  mutate(z_score = (follower_growth_pct - mean_growth) / sd_growth)

# calculate summary statistics
posts_with_z %>%
  group_by(posting_frequency_bin) %>%
  summarise(
    n_observations = n(),
    mean_z_score = mean(z_score, na.rm = TRUE),
    median_z_score = median(z_score, na.rm = TRUE)
  )

# A tibble: 5 × 4
  posting_frequency_bin n_observations mean_z_score median_z_score
  <fct>                          <int>        <dbl>          <dbl>
1 No Posts                     2422186      -0.0729       -0.262  
2 1-2 Posts                    1051055       0.0663       -0.129  
3 3-5 Posts                     616725       0.0936       -0.0840 
4 6-10 Posts                    277724       0.102        -0.0564 
5 10+ Posts                     256820       0.0812       -0.00813

This data suggests that there is a positive relationship between posting frequency and follower growth. Channels that post more frequently consistently gain more followers relative to their baseline growth.

It also suggests that there is a cost to not posting. Channels that don’t post at all under-perform their baseline growth rates. We can call this the “no post penalty”. Even posting once or twice a week results in a significant increase in follower growth compared to weeks in which channels post nothing.

The difference between moderate posting (3-5 posts) and more frequent posting (6-10 posts) is relatively small, and there could even be diminishing returns for channels that post more than 10 times per week. This suggests to me that finding a sustainable posting cadence matters more than just posting as much as possible.

Below I’ve plotted average Z-scores for each platform. For each one there is a penalty for not posting, and a strong positive correlation between the number of posts sent in a given week and the number of followers gained.

Fixed Effects Regression Model

A fixed effects regression model compares the same channel against itself over time, rather than comparing different channels to each other.

Instead of asking “Do channels that post more grow faster?” (which could be biased because high-posting channels might just be better at social media), fixed effects asks “When the same channel posts more vs. less, does it grow faster?” This within-channel comparison controls for all the unchanging characteristics that make channels different from each other.

The model essentially creates a separate baseline for each channel, then measures how posting frequency affects growth relative to that channel’s own average performance.

The key advantage is that we can make stronger causal claims about posting frequency. When we see that the same channel grows faster during weeks when it posts more, we can be more confident that posting is actually driving the growth rather than just being correlated with it.

# fit fixed effects model
fe_model <- feols(follower_growth_pct ~ posting_frequency_bin | channel_id, 
                   data = posts, 
                   cluster = "channel_id")
# summarise model
summary(fe_model)

OLS estimation, Dep. Var.: follower_growth_pct
Observations: 4,761,210
Fixed-effects: channel_id: 158,755
Standard-errors: Clustered (channel_id) 
                                Estimate Std. Error t value   Pr(>|t|)    
posting_frequency_bin1-2 Posts   7.87426    2.47737 3.17848 0.00148080 ** 
posting_frequency_bin3-5 Posts  10.47946    3.06451 3.41962 0.00062724 ***
posting_frequency_bin6-10 Posts 20.98967    6.42653 3.26610 0.00109064 ** 
posting_frequency_bin10+ Posts  31.87848    9.90608 3.21807 0.00129081 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 1,220.0     Adj. R2: 0.035287
                Within R2: 1.479e-5

These fixed effects model summary suggests that a clear, positive relationship exists between posting frequency and follower growth, with each level of posting activity driving higher follower growth.

The model shows that posting 1-2 times per week gains channels an average of around 8 additional followers compared to weeks in which they don’t post at all.

The benefits increase substantially with higher posting frequencies as well. Posting 3-5 times per week adds around 10 followers on average, while posting 6-10 times adds around 21. The most active channels posting 10+ times per week see the largest gains at 32 additional followers per week.

The progression shows larger follower gains up to the highest posting frequency. Each jump in posting activity seems to produce progressively larger benefits, for example going from 1-2 posts to 3-5 posts adds about 2.6 followers, while going from 6-10 posts to 10+ posts adds nearly 11 additional followers.

These results provide pretty clear benchmarks for a social media strategy, at least for these platforms. The data strongly supports the idea that consistent, frequent posting is a major driver of follower growth.

Conclusions

This analysis of 4.8 million channel-week observations across Facebook, Instagram, and X provides clear evidence that posting frequency directly impacts follower growth. Tthe Z-score analysis and fixed effects regression model tell the same story: posting more leads to better performance.

The data reveals several key insights to keep in mind. First, there’s a penalty for not posting. Channels that don’t post in a given week consistently underperform their baseline growth rates, which suggests that people and algorithms reward active accounts and penalize inactive ones.

Along those lines, any posting is way better than no posting. Even channels that post just 1-2 times per week see meaningful improvements over weeks when they don’t post at all. To me this suggests that consistency matters more than anything. There’s a real cost to not posting anything.

Third, the benefits of posting more appear to grow at higher posting frequencies rather than diminish. Channels posting 10+ times per week see the largest follower growth, adding an average of 32 additional followers compared to weeks with no posts.

The fixed effects model is particularly compelling because it compares each channel against itself over time, controlling for all the factors that make channels different from each other. When the same channel posts more frequently, it consistently grows faster. This gives us confidence that posting frequency is actually driving growth, not just correlated with it.

For creators, this analysis supports a strategy of frequent and consistent posting. While the optimal frequency varies by platform and audience, the data clearly shows that more active channels outperform less active ones. The question isn’t whether to post frequently - it’s how frequently you can post while maintaining quality and sustainability.

Engagement Rates

Next we’ll turn our attention to engagement rates. The SQL query below returns approximately 15.7 million posts and their engagement rates. It also includes baseline statistics for profiles like the average engagement rate, average number of posts sent per week, and the standard deviation of the number of posts sent per week.

sql <- "
  with profile_tenure as (
    -- calculate when each profile started posting and their tenure
    select 
      profile_id
      , user_id
      , profile_service
      , min(date(sent_at)) as first_post_date
      , max(date(sent_at)) as last_post_date
      , date_diff(current_date(), min(date(sent_at)), day) as tenure_days
      , count(distinct date(sent_at)) as active_days
      , count(distinct id) as total_posts
    from dbt_buffer.publish_updates
    where sent_at >= timestamp_sub(current_timestamp, interval 730 day)
    and profile_service not in ('pinterest', 'tiktok', 'youtube')
    group by 1, 2, 3
  )
  
  , qualified_profiles as (
      select distinct
        profile_id
      from profile_tenure
      where tenure_days >= 84 -- started posting at least 12 weeks ago
        and total_posts >= 20 -- minimum posts for statistical reliability
  )
  
  , profile_baseline as (
      select 
        profile_id
        , profile_service
        , user_id
        , approx_quantiles(case 
            when profile_service = 'instagram' 
            then (coalesce(likes, 0) + coalesce(comments, 0) + coalesce(shares, 0)) / 
            nullif(reach, 0) * 100
          else engagement_rate end, 2)[offset(1)] as baseline_engagement
        , count(*) as baseline_posts
    from dbt_buffer.publish_updates
    where sent_at >= timestamp_sub(current_timestamp, interval 365 day)
    and engagement_rate > 0
    and profile_service not in ('pinterest', 'tiktok', 'youtube')
    group by 1, 2, 3
    having count(*) >= 10 
  )
  
  , weekly_activity as (
      select 
        up.user_id
        , up.profile_id
        , up.profile_service
        , date_trunc(date(up.sent_at), week(monday)) as week_start
        , count(distinct up.id) as posts_in_week
        , approx_quantiles(case 
            when up.profile_service = 'instagram' 
            then (coalesce(up.likes, 0) + coalesce(up.comments, 0) + coalesce(up.shares, 0)) / nullif(up.reach, 0) * 100
            else up.engagement_rate end, 2)[offset(1)] as median_weekly_engagement
      from dbt_buffer.publish_updates as up
      inner join qualified_profiles as qp on up.profile_id = qp.profile_id
      where up.sent_at >= timestamp_sub(current_timestamp, interval 365 day)
      and up.engagement_rate > 0
      and up.profile_service not in ('pinterest', 'tiktok', 'youtube')
      group by 1, 2, 3, 4
  )
  
  -- calculate consistent time windows
  , time_windows as (
    select 
      profile_id
      , greatest(first_post_date, date_sub(current_date(), interval 365 day)) as analysis_start_date
      , current_date() as analysis_end_date
      , date_diff(current_date(), 
                greatest(first_post_date, date_sub(current_date(), interval 365 day)), 
                week) as total_possible_weeks
    from profile_tenure
  )

  select
    wa.user_id
    , wa.profile_id
    , wa.profile_service
    , pt.tenure_days
    , pt.first_post_date
    , pb.baseline_engagement
    
    -- posting frequency metrics
    , count(distinct wa.week_start) as active_weeks
    , sum(wa.posts_in_week) as total_posts
    , avg(wa.posts_in_week) as avg_posts_per_week
    , stddev(wa.posts_in_week) as posts_per_week_sd
    
    -- posting consistency metrics (using consistent time window)
    , tw.total_possible_weeks
    , case when tw.total_possible_weeks > 0 
           then count(distinct wa.week_start) / tw.total_possible_weeks
           else 0 end as pct_weeks_with_posts
    
    -- engagement metrics (using median)
    , approx_quantiles(wa.median_weekly_engagement, 2)[offset(1)] as median_engagement
    , approx_quantiles(wa.median_weekly_engagement, 2)[offset(1)] - pb.baseline_engagement as engagement_lift
    
  from weekly_activity as wa
  inner join profile_tenure as pt on wa.profile_id = pt.profile_id  
  inner join profile_baseline as pb on wa.profile_id = pb.profile_id
  inner join time_windows as tw on wa.profile_id = tw.profile_id
  where wa.week_start >= date_trunc(tw.analysis_start_date, week(monday))
  group by 1, 2, 3, 4, 5, 6, tw.total_possible_weeks
  having count(distinct wa.week_start) >= 4 -- at least 4 weeks of activity
"

# get data from BigQuery
posts <- bq_query(sql = sql)

Data Tidying

First we’ll remove outliers in the data and filter out profiles with fewer than 4 active weeks and 20 total posts. We’ll also only look at profiles that have been connected to Buffer for at least 6 months.

# function to calculate z-scores and remove outliers
remove_outliers <- function(data, columns, threshold = 3) {
  for(col in columns) {
    if(col %in% names(data)) {
      z_scores <- abs(scale(data[[col]])[,1])
      data <- data[z_scores < threshold, ]
    }
  }
  return(data)
}

# clean data and handle outliers
posts_clean <- posts %>%
  filter(
    total_posts >= 20,
    active_weeks >= 4,
    tenure_days >= 180, # 6 months
    !is.na(median_engagement)
  ) %>%
  # remove extreme outliers using z-scores
  remove_outliers(columns = c("avg_posts_per_week", "median_engagement"), threshold = 3)

For each platform and profile, we’ll plot the median engagement rate by the percentage of weeks in which posts were shared. The theory we would like to explore is that profiles that post in a higher percentage of weeks have higher median engagement rates.

# define consistency tiers
posts_clean <- posts_clean %>% 
  mutate(consistency_tier = case_when(
    pct_weeks_with_posts <= 0.33 ~ "Low",
    pct_weeks_with_posts <= 0.67 ~ "Medium",
    pct_weeks_with_posts > 0.67 ~ "High",
    TRUE ~ "Other"
  ))

# order factor levels
posts_clean$consistency_tier <- factor(posts_clean$consistency_tier, 
                                       levels = c("Low", "Medium", "High"))

# plot weeks with posts vs avg engagement
posts_clean %>% 
  mutate(pct_weeks_with_posts = ifelse(pct_weeks_with_posts > 1, 1, pct_weeks_with_posts)) %>% 
  group_by(profile_service, 
           consistency_tier = cut(pct_weeks_with_posts, seq(0, 1, 0.2))) %>% 
  summarise(med_engagement = median(median_engagement, na.rm = TRUE)) %>% 
  ggplot(aes(x = consistency_tier, y = med_engagement, fill = profile_service)) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~profile_service) +
  theme(legend.position = "none") +
  labs(x = "Percent of Weeks With Sent Posts", y = NULL,
       title = "Median Engagement Rate by Posting Consistency Rate")

The results are mixed. For several of the platforms there does seem to be a positive correlation between the proportion of weeks in which posts were sent and median engagement rate, but it’s only up to a certain point, around 50% of weeks. After a profile posts in at least 50% of eligible weeks the median engagement rates even seem to decline.