Social Media: Space for Self-Representation
The advent of social media sites, such as Facebook, Twitter and Instagram, has dramatically changed the way people and communities interact with one another. In recent years, the Oxford Dictionary has even added social media lingo into the dictionary, such as “unfriend”, “hashtag” and “tweetable”. The word “selfie”, popularised by social media sites, even won the Oxford Dictionary Word of the Year in 2013. In Singapore, up to 64% of the general population uses social media, prompting businesses and individuals to reach out to people via this platform. Jumping onto the bandwagon, politicians are also leveraging on social media to reach out to the masses.
Politicians use social media for their political campaigns as it allows them to reach out to both supporters and non-supporters relatively quickly, compared to traditional media. Social media also allows politicians to personalize their political campaigns, where the focus is placed on each individual politician instead of the political party. Social media sites represent “semi-public, semi-private spaces for self-representation”, allowing politicians to post about both their public roles and private lives. Through interviewing Norwegian politicians and content analysis of their Facebook and Twitter profiles, Enli and Skogerbø found that personal updates attracted more attention, as opposed to political statements.
To the best of our knowledge, no content analysis of the social media profiles of Singaporean politicians has been done. To this end, this project focuses solely on the Facebook profiles of Singapore ministers, as Facebook is commonly used by politicians to reach out to Singaporeans. Building upon Enli and Skogerbø’s finding, we hypothesize the following:
1. Defining engagement rate
Facebook defines the engagement rate for a post as the ratio of people who engaged with the post - by liking, reacting to, or commenting on it - over post reach, which is the number of people who saw the post.
However, analytics like post reach are only provided to owners of that particular Facebook page. Thus, a proxy was used to calculate engagement rate even without insight data. Assuming that the post reaches all followers of that page, the formula for engagement rate for our study was defined as the ratio of people who engaged with the post over the total number of followers for that page, as illustrated in the figure below. With this formula, it does not matter that Prime Minister (PM) Lee Hsien Loong has more than a million followers, but Minister for the Environment and Water Resources Masagos Zulkifli only has 12 thousand; we can still compare the two as engagement and follower count are expressed as a ratio.
Since engagement rate in the above formula is only for a single post, our next step was to come up with a metric that covers all posts by the same person. Besides number of followers, we had to consider that some politicians might be more active than others on social media. For instance, PM Lee had close to 500 posts in 2016, while Mr Masagos had only half of that number of posts. With that in mind, we came up with the mean of engagement rate of all posts by summing up the engagement for all posts and then dividing this number by the total number of posts by the same politician. This accounts for the different number of posts by different politicians. From this, we then have:
The mean engagement rate (MER) of politicians is still an unfair comparison as politicians who post more personal posts had spent more time doing so. Let’s say two politicians have the same engagement rate, but Politician A spends more time on personal posts than Politician B. The rate of return on time spent for each politician is then different, with Politician B having a higher rate of return as he had spent a lesser amount of time on personal posts but has the same engagement rate as A. Hence, we also have to consider the time spent on personal posts in our metric to be able to make a fair comparison between politicians. We do not know how much time each politician spends on a post but by assuming that an equal amount of time is spent on every post, we can express proportion of time spent on a particular category as the ratio of number of posts in that category over the total posts.
Putting the MER and proportion of time spent together, we now have the MER per unit time (MER/time), which we define as MER divided by the proportion of time spent on a particular category. Specifically, since we are looking at whether a larger percentage of personal posts equates to a higher engagement, we are interested in the MER divided by the proportion of time spent on personal posts. We now have our metric that is ready to be used:
2. Obtaining data
After deriving the metric, we had a clear understanding of the data we needed to obtain from politician’s Facebook pages. Facebook has developed tools known as “Facebook Graph APIs” for app developers to be able to extract useful data from Facebook pages programmatically and with relative ease.
As such, the next step in our project was to write the relevant code to obtain the following data from Facebook: text status of post, timestamp of post, and number of shares, comments, likes and reaction (“Love”, “Haha”, “Wow”, “Sad” and “Angry”).
We decided to work with Python programming language and Jupyter Notebook, a popular tool for writing Python code. Python was chosen primarily due to the useful libraries that come in handy for this project.
First step was to import the following libraries:
Most of them are standard python packages used for parsing, storing and manipulating data. The one that is specific to obtaining data off Facebook would be the “Facebook” module. While it is possible to use the Graph API without it, the module makes it much simpler and neater.
Then we extracted the data we need from each politician’s page using the following code:
In , the ACCESS_TOKEN string needs to be keyed in manually. This token is required for a person to obtain any information from Facebook, and the the token is restricted by privacy rights. We generated a temporary access token using Facebook Graph Explorer and since all the data we wanted to extract are from public pages, there were no privacy issues. The END_DATE needs to be specified which sets how far back in time we want the code to pull posts from. Posts are ordered from newest to oldest so by setting a date in the past, in this case 1st Jan 2016, we will be pulling post from data from 1st Jan 2016 till today.
In , we need to manually state the identity (“id”) of the politician’s page we intend to scrape, which is the unique identifier for a politician’s Facebook page. If we wanted to pull for instance, K Shanmugam’s page, we will use id = ‘k.shanmugam.page/posts?limit=100’. The “id” can be identified from the page URL as seen from the image below:
The code so far only returns the latest 100 posts (maximum per API call). Since our aim is to look at posts from a given END_DATE, we will continue to loop through the code, obtaining 100 posts at a time, till the END_DATE is reached.
The code will print the number of posts stored and the date of the oldest post. Subsequently, it will pull the relevant data (likes, shares, comments etc.) from these posts and store them temporarily in a list.
Finally, the code in  organises the data in the list into a csv file that we can use to draw analyses. This whole code was run for each politician’s page to obtain all data we need. The code is publicly available on GitHub here and can be cloned to a local repository for anyone to view, should they want to expand on our project.
3. Categorising data
After obtaining the required information using the code, we manually categorised the posts into two categories - ‘personal’ and ‘political’. An example of a personal post is:
And an example of a political post is this:
However, some of the posts are somewhat ambiguous, as it comes across as personal but at the same time, overlaps with the minister’s portfolio or job scope. Here is an example:
This post is considered ambiguous because although Ms Fu seems to be sharing her achievements on completing a run, she uses this opportunity to encourage Singaporeans to engage in a more active lifestyle as well as raise awareness on supporting charitable organisations. This overlaps with her role as the Minister for Culture, Community and Youth.
So instead of having only two categories, we are now shelving each post into one of three categories: ‘personal’, ‘political’ and ‘ambiguous’. It would be interesting to see if the ambiguous posts showed any trends, and also if politicians employ ambiguity as a strategy to garner more attention.
4. Visualising data
Finally, we used Tableau to visualise and organise our data. Click on the next page to see what we found out!
What did we find out?
The presence of ambiguous posts complicated our analysis. It would not be sufficient to look solely at the relationship between the Mean Engagement Rate per Unit Time against the percentage of personal posts as now, there is some uncertainty involved; an ambiguous post could be seen as either political or personal.
A fair way to identify the overall trend would be to use a sensitivity analysis. Sensitivity analysis refers to a technique of varying a variable and looking at how it impacts another, holding everything else constant. The level of this impact would signal how sensitive the relationship between the variables are; significant impact would mean highly sensitive and minimal impact would mean insensitive.
In our case, since the uncertainty lies within the classification of the ambiguous post, we decided to vary it to both extremes, i.e. one case would be when all ambiguous posts are taken to be political and the other case would be when all ambiguous posts are taken to be personal. These cases formed the upper and lower bounds respectively. We also found the midpoint between these boundaries by categorising half the ambiguous posts as personal the other half as political. We then visualised the data in the chart below (interactive version available).
Fortunately, the upper and lower boundaries are not wide apart and all three cases (upper, middle, lower) follow a similar trend. Consequently, this would mean that our model is reasonably robust and the uncertainty of ambiguous posts do not undermine the trend identified.
Having established this, there seems to be an inverse relationship between the percentage of personal posts and MER/time. This is contrary to our initial hypothesis and the Norwegian study.
Why is this so?
The trend that we observed from our data appears to be the opposite of what the Norwegian research has shown. One of the possible reasons could be due to the cultural differences between Norway and Singapore, which we can analyse using the Hofstede Theory of Cultural Dimensions.
In our analyses, we are focusing specifically on two dimensions: power distance and masculinity.
There is a relatively higher power distance in Singapore as compared to in Norway. What this means is that in the Singaporean society people tend to be more hierarchical and less personal with their superiors than in the Norwegian society. How we can relate this back to our research is that Singaporeans could possibly be viewing politicians as positions of power – positions that they would not like to get more personal with. Hence, this could contribute to the trend that Singaporean politicians should post political posts more so than personal posts.
Singaporeans also appear to be more ‘masculine’ - driven by competition - as compared to Norwegians. Being more results-driven could explain why Singaporean politicians appear to do better when they post more political posts as compared to more personal posts. Perhaps political posts convey the message that something is being done with regards to their portfolio, whereas personal posts may convey a sense that politicians are not working for actual change.
In conclusion, the differences in culture between Singaporeans and their Norwegian counterparts could have caused our results to have an opposite trend from our hypothesis, which was based off the Norwegian study.
While the primary goal of our project was to test the initial hypothesis we had made, we chanced upon several interesting trends/patterns/observations as we analysed the data we had collected. We decided to visualise and share some of the interesting ones! Feel free to dive further into any of these observations in your leisure time. (Don’t forget to hit us up if you find something cool, we’d love to know more about it!)
1. Many politicians seem to have scheduled posts
There appears to be a higher frequency of posts by politicians at “:00” and “:30” min of each hour (e.g. 12:00pm, 1:00pm, 2:00pm and 12:30pm, 1:30pm, 2.30pm), suggesting that these posts were scheduled to be posted at these times. This can be seen in the chart below (“Number of Posts Published over Time (Minute)”). Facebook has the feature to do so, and this might be a way for politicians to update their Facebook pages and engage with their followers in the midst of their busy work schedule. They can plan out the posts ahead of time when they are free and have them posted at appropriate times.
The chart in red, “Number of Posts Published over Time (Hour)”, shows the post frequency distributed over the day. While there seems to be relatively higher frequency of posts towards the end of the day, the overall variance throughout the day is not particularly significant. Social media marketers generally believe there are “prime times” to post content on Facebook to maximise post engagement. However, the data seems to suggest Singaporean politicians are not exactly following such techniques.
2. MER/Time vs. % ambiguous posts - no observed trend!
Social media allows politicians to connect to the masses at a more personal level, as compared to traditional sources of media. Politicians could possibly use this to their advantage and impart their views on a larger social issue by intertwining sociopolitical and personal views within a post. We observed such posts during our analysis and they were labelled as ambiguous. So we decided to test an alternative hypothesis: the strategy of posting an ambiguous message (ie. message with both personal and political content) boosts their ability to garner more engagement as these posts break down and present complex issues to the general public in a relatable way. That is, the higher the percentage of ambiguous posts, the higher the MER per unit time.
Interestingly, we did not see any trends when we visualised the data - MER per unit time against the percentage of ambiguous posts. As such, we are inclined to believe that politicians (at least, most of them) do not consciously use this as a strategy to reach out to the public. Posting of ambiguous posts may not be an effective strategy in Singapore. Perhaps Singaporeans prefer to view posts about ministers’ portfolios rather than the ministers’ thoughts on recent sociopolitical issues.
3. Frequency of posts across the year (by month)
As seen from the chart below, there was an increase in the number of Facebook posts by politicians in August. This could be attributed primarily to National Day, the 2016 Rio Olympics and the passing of Mr SR Nathan.
Most, if not all, ministers talked about the National Day parade (preparations for National Day parade, attending the parade on the actual day, celebrations in their various constituencies etc.). In the 2016 Rio Summer Olympics, Joseph Schooling won Singapore’s first Olympic gold medal and as expected, ministers expressed their national pride on Facebook. The passing of Mr SR Nathan further sparked a series of posts by ministers who expressed their condolences, shared their memories and paid tribute to him. These factors all contributed to August having the highest number of total posts last year.
4. PM Lee Hsien Loong's Public Relations (PR) Team
An interesting observation we made was that PM Lee Hsien Loong’s Facebook page has a combination of posts by him and his PR team that manages some of his social media content. When a post is published by the PR team, they will address PM Lee Hsien Loong in third person. In contrast, PM Lee Hsien Loong uses first person to address himself when publishing personally.
Post by PM Lee Hsien Loong’s PR team:
Post by PM Lee Hsien Loong:
5. Ministers with community-related portfolios had a higher percentage of ambiguous posts
We defined ambiguous posts as posts which relate to both the minister’s private lives and their job scope. Ministers with community-related portfolios tend to have a higher percentage of ambiguous posts due a larger overlap between their personal lives and jobs.
Minister Grace Fu, who is the Minister of Culture, Community and Youth, had the highest percentage of ambiguous posts. She regularly posted about sports and arts festivals in Singapore, describing her enthusiasm for these activities and encouraging the general public to be more active in participating in such community events. While these might be her hobbies, she could have also posted these in her capacity as the Minister of Culture, Community and Youth.
In contrast, ministers without community-related portfolios have a lower percentage of ambiguous posts. For example, Minister Chan Chun Sing is the Minister for Prime Minister’s Office (PMO), which oversees various issues of national importance, such as corruption, elections and policies, and manages the various ministries. As such, he has one of the lowest percentage of ambiguous posts - it is relatively easy to tell the difference between a political and personal post for his portfolio.
Should politicians share their personal lives on Facebook? If their goal is to maximise engagement, our analysis suggests that they shouldn’t. To effectively reach out to the masses, politicians should focus more on sharing posts with political content rather than one that is purely personal. Nevertheless, Facebook is still a space for everyone to share their personal experiences and emotions; perhaps the best place to post about their favourite #durianparty. Would you really want PM Lee Hsien Loong to stop posting about #guesswhere or #jalanjalan? After all, these posts show a more human side of him which might have been the very reason some decided to follow his page in the first place. While personal posts may not directly translate to greater engagement, it might play a role in attracting more followers. Over time, with the increasing presence of social media in the average Singaporeans’ life, the observed trends may change; perhaps closer to what was seen in Norway.