Conservative MP Twitter power rankings: the top 50
Rob Sutton is an incoming junior doctor in Wales and a former Parliamentary staffer. He is a recent graduate of the University of Oxford Medical School.
Amongst the social media giants, Twitter is the primary battleground for political discourse. It’s also one of the key avenues by which MPs convey their message, and has near-universal uptake by members in the current House of Commons.
The effectiveness with which Twitter is utilised varies considerably between MPs, but it is difficult to compare like-for-like. How does one take into account the differences between, for instance, a freshman MP and a veteran Cabinet member? Length of service in Parliament and ministerial rank give a considerable advantage when building a following.
In this article, I have compiled a power ranking of MPs in the current Parliament, with the top 50 shown in the chart above. The MP’s follower count was adjusted by factoring in their previous experience, to better reflect the strength of their following and their success at engagement on the platform.
Being Twitter-savvy is about more than just a high follower count: any Secretary of State can achieve this just by virtue of the media exposure their office brings. Building a Twitter following based on thoughtful commentary and authentic engagement requires skill ,and can be achieved by members across all Parliamentary intakes and ranks of Government.
Though the top 10 is still dominated by MPs holding senior ministerial offices, the composition of the list beyond it is far more variable. A number of prominent backbenchers are in the top 20, and four members from the 2019 intake make the top 50, beating longer-serving and higher-ranked colleagues.
I hope that this list serves as recognition of the skill and contribution by Conservative members to public debate and engagement, beyond ministerial duties which so often dominate any mention in the media.
Building a model of Twitter power rankings
Success is judged by number of followers, with higher follower counts indicating greater influence on Twitter. The follower count was adjusted using three key parameters:
Higher values for each of these would be expected to contribute to a higher follower count. I built a model using the open-source Scikit-Learn package, and fitted it to data from the current Parliament.
The model was then used to predict how many followers a given MP might expect to have based on these three factors. The steps taken to produce a final “Twitter power score” were thus as follows:
The final step of taking a logarithm means it is easier to compare between MPs without those who have very high follower counts (such as Boris Johnson) making the data difficult to compare, but it does not affect the order of the ranking.
Compiling the data
Having decided which factors to correct the model for, I collected the required information. All three factors were easy to find reliable sources for. The Twitter page for each MP displays the date the account was created, and the Parliamentary website provides the date of their first election to Parliament and previous government posts.
Members who are newly returned to the backbenches following governmental duties (such as Sajid Javid and Jeremy Hunt) are scored at their highest government rank since 2010 to recognise this. I was able to find the Twitter accounts and required information for 319 Conservative MPs who were included in this ranking.
To build a model based on this data required incorporating the highest government rank numerically. To do this, I assigned scores according to their rank. These grades recognised their relative seniority and media exposure associated with the office, with higher scores assigned to more senior positions:
When assigning these values, I considered the typical sizes of follower counts of MPs in each category. When comparing Secretaries of States to Ministers of State, the median follower count is around twice the size, but the mean follower count is around eight times the size, as a handful of very large follower count skews the results upwards.
Deciding on weightings requires a (somewhat arbitrary) decision as to which measures to use when comparing between groups, and the scores I decided on were ultimately chosen as a compromise across these different measures, which produced stable results when used in the model.
It is also worth explaining why Prime Ministers are grouped with Secretaries of State, despite the far higher media exposure and seniority of their post. When deciding on the respective weighting for different levels of government post, a sufficiently large pool of MPs was needed to produce a meaningful comparison. The only data points for comparison of Prime Ministers are Boris Johnson and Theresa May, so it is difficult to give them their own weighting without it being either unreliable or arbitrary.
While grouping them with Secretaries of State and other senior positions might be perceived as giving them an unfair advantage in the weighting, I felt it justified given these challenges in determining the “fair” weight to assign them. With this done, I had three parameters for each MP on which to build a model to calculate the expected number of Twitter followers.
Calculating the number of expected Twitter followers
I built a model to calculate the expected number of Twitter followers using the Scikit-Learn, a popular machine learning package in the Python programming language. The model used multiple linear regression to fit the input parameters to the known follower count.
The input data was prepared by removing extreme high outliers in the data which skewed the fit toward high numbers and away from the vast majority of MPs before fitting. Once fitted, an “expected value” of Twitter followers could be calculated for each MP, based on the year of their first election to parliament, the number of years on Twitter and their highest government rank since 2010.
Including more parameters increases the ability of the model to describe the difference between MPs’ follower counts (the variability). By increasing the number of input variables included in the model, more of the variability is captured:
These three variables are therefore responsible for almost half of the variation between MPs in their follower counts. The remainder of the variability is likely due to a range of factors which the model does not include, of which the MP’s Twitter-savviness is of particular interest to us. I discuss these factors further below.
Limitations in the model
There are multiple other parameters which could be included in future iterations which I did not include in this model. In particular:
Another limitation was not accounting for the perceived relative importance of various governmental departments: a Great Office of State or Prime Minister is scored the same as any other Secretary of State. The difficulties involved in ranking governmental departments were beyond this first model. The length of service in a given government post was also not considered.
Finally, the choice of model to fit the data may not be the optimal choice. Multiple linear regression assumes, per the name, that the distribution is linear. But the large outliers might be better described by a power law or Pareto distribution, or the non-linearities of a neural network.
During next week, ConservativeHome will produce profiles of six individual MPs who have performed notably well in the power rankings, and who reflect the contributions brought by members beyond their ministerial duties, if they have any.