(101 Introduction To Social Analytics+ 201 Boolean Mastery+301 Data Analysis And Insights Generation)
301.3 Qualitative Analysis
Please watch the video, read the lesson below and take a quiz at the end of this lesson (bottom of this page).
Once we get an overview on the dataset from quantitative analysis, the next step will be to explain why these trends are happening. Unlike quantitative analysis which focuses on the numbers and its breakdown, qualitative analysis takes a more subjective view into research – which provides the reasoning behind themes and human behavior from social media. This includes clustering all conversations into relevant “buckets of information” to understand the different layers of social conversations.
While deep-diving straight into each and every conversation may require considerable effort, a good place to start is to get a sense of the overarching themes using some built-in functionalities from Radarr
These include analysis such as the Word Cloud and Top Topics under the Insights section.
301.3.1 Common Methodologies from Qualitative Analysis
Diagram 301.3.1a: Word Cloud to show the most frequently mentioned keywords on conversations around snacking from Radarr.
Unlike a frequency table that lists down the keywords with most mentions, Word Cloud is a visual way of depicting cited keywords across topics – which gives an indication of the underlying themes at a glance. With the example above, we can infer a couple of emerging themes from the top keywords. In addition to the specific snacks that people are interested in, most of the conversations were centered around recipes (e.g. cream, milk, egg), flavors (e.g. japanese, cheese) and purchase intent (e.g. purchase, address, shop).
Do note that on Radarr, keywords in blue are the most frequently mentioned, followed by those in green and yellow.
Each of the keyword on the Word Cloud can be further explored by
- Viewing the raw conversations using the specific keywords (right click > View Conversations > select); or
- Viewing the sub-cloud of specific keywords to further analyse associated keywords (left click on keyword)
The emerging themes are then used as inspiration to do a qualitative deep-dive to get more granular and nuanced insights from the dataset.
|Note: While Word Cloud typically depicts keywords by frequency, Radarr can also visualise keywords by Impact Score, which will instead showcase the top phrases used in the most engaged-with conversations and posts online.|
Diagram 301.3.1c: Topic Clustering to show prominent groups of conversations around Snacking from Radarr.
Although Word Clouds are great with helping users understand the key terms that are being used within certain topics, social media conversations doesn’t always mean the same thing when the exact same words are used – with evolving internet lingos and writing styles that differ across different countries and demographics.
This is where theme clustering would help users make sense of what are the trends, discussions and events that are happening online. The Top Topic function on Radarr does just that – allowing users to have a birds eye view on what are the main subject of discussion without going down to the specificity of the keywords.
Diagram 301.3.1d: Exploring the functionalities of Top Topic [.gif]
Each of the clusters on the Top Topic can be further explored by
- Viewing the raw conversations within the specific theme (right click > View Conversations > select); or
- Viewing the sub-themes of specific main themes to further analyse associated posts (left click on theme)
|Note: The topic clusters on Radarr are also color-coded to reflect the general sentiment within each theme. Green represents positive sentiment, red represents negative sentiment, and grey for neutral sentiment. The shade of color signifies the intensity of sentiment, e.g. dark red means that most posts within the theme are of negative sentiment, light red indicates a closer balance with the neutral sentiment.|
Similar to the Word Cloud, the emerging themes are then used as inspiration to do a qualitative deep-dive to get more granular and nuanced insights from the dataset. The choice between Word Cloud and Top Topic is dependent on the type of analysis and insights that the user is trying to derive from the qualitative deep dive, and both can be used collectively to improve the robustness of the research process.
301.3.2 Deep-Diving Into Qualitative Data
In addition to designing the research and data optimization, one of the most imperative (and exciting) parts of being a social analyst is in breaking down the qualitative data in a way that will help answer your initial research questions. Not to say that quantitative breakdown and AI-powered qualitative analysis is not important to the research process, but since these automated insights are readily available on the social listening tool, qualitative deep-dive is where analysts are best able to integrate their local/industry knowledge and understanding to the dataset in crafting the “why” behind these trends.
As covered in the A-Z of a Social Analyst in the 101 course, the qualitative deep-dive is primarily done through 5 key steps:
- Determine the specific topics to explore
- Review and explore data
- Creating initial themes and tagging dataset
- Review themes and clustering into macro-themes
- Integrating the findings with quantitative insights
Step 1: Determine the specific topics to explore
Depending on the initial objective and hypothesis, the topics of exploration can be both top-down (i.e. having specific hypotheses to be tested) and bottom-up (i.e. evaluating the data without any preconceived notions).
Top-down approaches are usually led by certain hypotheses. Although these may limit the coverage of analysis, this approach is a lot more targeted, whereby resources can be specified to capacity for action. Examples of research questions suited for this approach include:
- Where should I focus on marketing efforts, between Facebook and Twitter, for gaming consumers in SEA?
- Which of my brand proposition resonates best with consumers online?
- How serious is the online backlash from launching a LGBTQ-friendly programme in Asia?
On the other hand, bottom-down approaches are more exploratory, where research delves into a wider scope for analysis without preset restrictions. While this is more comprehensive, the findings may not always be within the capacity for action for your function (e.g. a marketing team cannot do much to change product development, but they do have the ability to shape the messaging to control perception towards existing products). Examples of research questions suited for this approach include:
- Which are the most popular snacking flavors in Singapore?
- What are the main pillars of sustainability that millennials care about?
- What is the main motivator for netizens to participate in online challenges?
While preliminary research typically applies the bottom-up approach, the application is very much dependent on the business need and ability for the different functions to action on these findings. For instance, if your brand is looking to launch a new flavor that is in line with health and fitness trends, there is little value in broadening your exploration to every flavor profile being discussed online. It may be more effective to shortlist some health-related flavors based on your industry knowledge (forming hypothesis), and subsequently reviewing (i) which are the most popular flavors and (ii) what are the emerging themes behind each flavor to determine which has the highest potential for a successful launch.
Knowing the specific questions you want answered will help you define the group of data you wish to analyse – which refines the quality of analysis.
Diagram 301.3.2a: Chart depicting topic selection as a subset of overall data collected from social listening tool.
Step 2: Review and explore data
Once the relevant data has been extracted by (i) creating topics and subtopics or (ii) filtering the relevant data using booleans on the search bar, it is necessary to review the relevance of the dataset before you start analysing the data.
|Note: Creation of topics and subtopics, as well as filtering of relevant data on the search bar was covered in 201: Boolean Mastery. Please refer to the course syllabus should you need a refresher for this section.|
A quick way to review data relevancy is to start by eyeballing the dataset. If the data is clean enough that you are able to pick up potential answers to the research questions, you can proceed to noting down these emerging themes for further analysis. However, should you find that the booleans that were used to filter these data can be optimized, this is when you can follow the Data Optimization process to clean up the dataset.
Diagram 301.3.2b: Process flow for reviewing and exploring data
|Note: Data Optimization process was covered in 101: Introduction to Social Analytics. Please refer to the course syllabus should you need a refresher for this section.|
Now that the dataset is clean and you have identified some preliminary themes, you can start doing a qualitative deep-dive by analysing each datapoint. This is the part of research where the analyst is able to discover the prominence and uncover the deeper nuances behind each theme.
Sampling may be required for a dataset that is too extensive for a 100% coverage. Consider a data pull of conversations around general snacking – which can easily have hundreds of thousands of posts in a month. In these cases, the analyst needs to make a judgment call of whether to do a random sampling, and if so at what percentage that will make most sense for the use case.
|As a rule of thumb, there should be at least a n=300 or 10% sampling – whichever possible, for a sufficient scan of conversation. The sampling should only be done after thorough data optimization to ensure that the sample is from a dataset of relevant data points.|
Step 3: Creating initial themes and tagging dataset
Based on the preliminary themes identified in step 2, the data can now be tagged based on how well it fits into each of the themes identified. This essentially means to allocate a value to each datapoint – depending on the content and context of the conversation.
Data tagging can be done via two methods:
- Using the Collection function on Radarr
- Downloading the dataset for manual tagging
Using the Collection function on Radarr
- Log into Radarr.
- Access the topic of interest by selecting the relevant filters under the Conversations tab.
- Hover over post to access the different options.
- Select “Add to Collection”. Each collection serves as a board for you to collect posts that are relevant to a certain theme – so that the original data can be easily accessible after tagging.
- If this is the first post that is being added to the theme, input the name of the theme under “New Collection”
- If this is a post to be added into an existing theme, select the drop down icon next to “Collection”, where previously added themes will be displayed.
- Once all that is done, click “Save”
- To access the Collections, head over to the left tab > Collaborate > Collections > Select “Own” and you will be able to access the different Collection created.
Diagram 301.3.2c: Steps to create a Collection on Radarr [.gif]
|Note: By default, Radarr will save the Collection under your “Own” view, which means that other users will not be able to access the Collection you created. For collaboration between team members, a “Team” will need to be set up on the Admin Panel for shared views across different users on Radarr.|
Downloading the dataset for manual tagging
While the Collection function is perfect for those who prefer a more visual interface to view the dataset, an alternative is possible for those who work better with spreadsheets. To download the dataset in .xlsx or .csv:
- Log into Radarr.
- Access the topic of interest by selecting the relevant filters under the Conversations tab.
- Click on “Download” at the top panel of the page.
- Select the preferred file format for download between .xlsx or .csv.
- A popup will appear, allowing selection of different variables for the download. “All” is selected by default.
- Click on “Download” and you will receive the download link via your Radarr registered email address.
Diagram 301.3.2e: Sample data downloaded via Radarr. Specific themes can be manually tagged to each datapoint in an additional column (e.g. column AG and AH as shown above).
|It is important to bear in mind that the current data tagging is based on preliminary theme identification, which means that you might still come across new themes that are relevant as you deep-dive into the dataset. Just create add on the theme as a new item to be tagged and you’re good to go!|
Step 4: Review themes and clustering into macro-themes
Now that the themes are finalized and data is tagged, it is time to review the themes and understand the underlying explanation behind this occurrence. Take for instance the themes that you have found when analysing data around health and fitness during the lockdown due to the COVID-19 pandemic in Singapore.
Based on the analysis, you may have noticed that most people were talking about these 12 activities:
● Home Workout
These themes may seem rather discrete for now, and there is not much insight that is actionable apart from understanding how these activities fare in terms of its popularity between one another. However, there are a couple of ways that we can cluster these into meaningful macro-themes that will help us better understand the motivation behind interest towards these activities.
The analyst can come up with several explanations based on the industry and cultural understanding of the local residents, for instance:
- Are the activities mostly outdoors or indoors?
- Due to fear of exposure to COVID-19 by going outdoors
- Are the activities mostly done individually or in a group setting?
- Due to prolonged period of staying at home and not socializing
- Are the activities mostly free or paid?
- Due to Singaporeans’ appreciation for a free and good deals
Based on the findings from these topic clusters, the analyst will be able to better quantify the prominence of these hypotheses – and subsequently come up with better insights that will help drive actionable recommendations for their business functions.
Diagram 301.3.2f: Different possibilities of cluster based on industry and cultural understanding of local residents [.gif]
Step 5: Integrating the findings with quantitative insights
The final step in the qualitative deep-dive will be to reconcile these findings with the quantitative findings based on the type of insights that you would need to generate. This will be covered in the next section, 301.3 Insights Generation: Making Sense of Quantitative and Qualitative Analysis