Meetup Engagement: What Makes a Successful Meetup Group
The skills we demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Whether you’re engaging with the public to sell a product or service, looking to improve your skill in a specific subject, looking for a community of like-minded individuals, or simply looking to make new friends, attending events is an essential component of building community.
However, event planning is complicated and costly, the event planning industry in the U.S. generated $4 billion in revenue in 2019 (statista.com). Everyone’s been to a party where no one or very few people show up. Wouldn’t it be nice to know what specific things can be done to increase the likelihood an event will be successful?
I sought to understand what characteristics successful groups on Meetup.com share. I explored this by examining data found on Kaggle.com related to meetup events in Nashville, Tennessee during November 2015 - October 2017. Briefly, Meetup.com is a website that allows members to form groups and for its members to search groups and locate those of interest. Meetup.com groups, as the name implies, are geared towards those that want to “meet up.” The website offers various ways to search or filter groups to find groups of interest. Groups can be private or public.
The data set contained 24.6k members, 602 groups, and 19k events. The data included RSVP data for each event and identified members that RSVPed for each event. Each group was listed under a category (e.g. socializing, photography, tech, etc.).
A subset of these data were extracted and used for analysis in order to reduce the data size and control for scenarios that might have artificially skewed the data. For example, the group creation date was not part of the data provided. As such, including data on groups that were created near the end of the time period (Nov 2015 - Oct 2017) would highly skew the data toward a low number of events and low membership.
I took the most recent year of data and only retained groups with an event at least a year before the end of the data set. In other words, I used the first event date in the data set as a proxy for group creation date and excluded those that were created during the time period in question (i.e. the last year of data). A full year of data was important to capture a representative sample of any groups that experience seasonality. The subset of data used included 13.5k members, 349 groups, and 7.9k events.
The most common day for events to be scheduled was Saturday and the least was Monday.
The category of group with the most number of events was Career and Business. However, when looking at a box plot of the same data, it was clear that there are some outliers in this category. Additionally, Career and Business was the second highest in terms of the number of groups. When looking at the number of events per group Dancing had the highest median number of events per group. Career and Business fell to the middle of the pack in this graph.
Many of the Career and Business groups had an extremely small number of events. This may indicate that these groups were not created with the intent to grow a community, but instead may have been intended for another purpose such as a singular networking event, or a college class project, etc.
I wanted to see if the number of members had a relationship with the number of events. Plotting number of events vs number of members did not reveal an obvious relationship. However, exploring some of the extreme examples there was an investment group that had a low number of members, but a large number of events. In contrast, a social group had a high number of members and only 1 event.
These examples point to the importance of segmentation (i.e. the investment club most likely limits membership and has a vetting process). And may indicate outliers impacted by events not represented in the data. The social club had over 4k members, but only 1 event. This may indicate that the group was dissolved during the time period of these data and their last event was captured in the data set.
The exploratory data analysis completed on these data revealed that Saturday is the most common day for an event, and that Monday was the least common. The number of events within each category was highly impacted by the number of groups in each category. When considering the number of groups within each category, their order changed considerably as compared to the total number of events in that category. In other words, Career and Business groups didn’t necessarily have many groups with repeat events, but instead just had more groups with limited events.
Initially, using the included categories to look for trends in specific types of groups and popularity seemed promising. However, it became clear that groups were placed into single categories and that in reality many groups were part of multiple categories. For example, a group called “20's & 30's Women looking for girlfriends” was listed under the category “Socializing,” when clearly it could also be listed under “Singles” or “LGBT.”
The main future direction for these data is further segmentation. It would be useful to identify the goal of each group and their intended audience. This would enable filtering out the groups which do not intend to grow a large network out of the data.
Additionally, identifying the types of events that the group is hosting may be a cleaner, more reliable way to categorize groups rather than relying on how they self-categorize. Refreshing the membership numbers to those that have attended an event in the last year to remove abandoned accounts. As well as removing groups that may artificially inflate or suppress membership will reduce noise and allow for a more comprehensive analysis of the data.
Ideally I would be able to distinguish which groups are engaging their audiences in additional ways beyond events. And which groups may be created from an existing group external to Meetup.com.