00:00:06 Introduction

Thanks Faye, and thanks to everyone for joining us. Before we get to the webinar, I'd like to just take a few minutes to tell you about ActivityInfo. BeDataDriven is the company that we work for, and it was founded with the mission to help social impact organizations do their work more effectively through the use of data. The way that we're delivering on that mission is by developing the ActivityInfo platform and helping organizations implement it.

ActivityInfo is an end-to-end solution for M&E data management that's built on a relational data model. It really supports collection, management, and analysis of that data, and it has all the tools that you need to manage data across the entire data lifecycle in a single system. It provides web and mobile applications for data collection in the field or in the office in a disconnected or connected environment. It has intuitive built-in tools and an expression language that helps validate data at the point of entry and clean data sets inputted from other systems. It really helps with harmonizing.

It is based on a robust cloud system architecture that ensures that data stored in the system is always available, it's secure, and there's no need to manage any of the hardware or cloud environments yourself. You just open the app and get going. There are also data management capabilities to help organize the data and to provide access to those who are permitted to access that data. Lastly, and it is the subject of our discussion today, there are really powerful analytic and visualization tools that help you extract value from your M&E data to help with decision-making.

00:02:48 Importance of data visualization

So what we will see today, basically we will dive into data visualization. We will start by outlining the importance of data visualization. Then we will take a deeper dive on the principles of good data visualization. What does it mean to understand our audience and the purpose, choosing the right chart, what are some best practices for clarity and consistency, color, and to which extent the color plays a role, or the font size, etc. Then we will see some examples by identifying good and bad visualization for very specific and commonly used charts. At the end, we will go through a few examples; we will analyze real-world examples with the use of ActivityInfo to showcase to you how data visualization looks like in practice.

In terms of introductions, I found this really nice phrase. It was a book that was written by Henry Hubbard of the National Bureau of Standards. The book is called The Magic of Graphs. What strikes me is the fact that actually a curve, he mentions, can inform the mind, awaken the imagination, and convince. And it's one of the most important points for me because indeed data visualization is a very powerful means of communication. We can communicate a very simple message, but also a very complex message to different kind of audiences. The charts carry the message itself. So they can be considered a universal language that actually conveys the information directly to the mind.

Having said that, the importance for me of data visualization starts from the fact that data visualization enables us to explore. We are trying to explore and understand frequent patterns and trends with our known data. Also, it helps us explain; there is something in our data that we need to communicate to our audience. We can choose different graphs depending on the question that we have, on the target audience, on the data that we have collected. And at the end of the day, it makes us think effectively. Good data visualization allows us to reason and think effectively about our data. Example given, if we see numerical data in a table, we may be able to find the trend, but it will take us a while to identify the trend. But if we plot that data into a nice line chart or scatter plot, then the trend becomes immediately clear to our minds.

00:06:29 Understanding the target audience and purpose

Now, what is the starting point, especially when we want to visualize data for the purpose of the humanitarian and development world, for the purpose of monitoring and evaluation? The key is to understand who is the target audience: inside the organization, outside of the organization, people working in the field, people working in the office, or people working in the HQ. Second, what we are trying to communicate; what is the main question that we want to answer? We want to see how many services we have delivered, if the services vary compared to different partners or different regions. So it's the "what" also that drives the choice of the chart and how at the end of the day we present information. Then it comes the choice of the appropriate visual.

Consider how this starting point is actually reflected in our terminology. Frequently, or the most common scenario, is that we start with the M&E plan. The M&E plan is kind of the common dictionary, the common vocabulary within the M&E system. There, we identify the data that we need. We identify the type of analysis. So we specify the type of calculation: do we count? Do we divide? We identify the report that we need. Who is going to use that indicator? We need it for the donor reporting? We need it because it's important for the field coordinators? We need it because at the end of the day we gather this information at the HQ on a yearly basis? So there, it's actually the core, the heart of the identification of the correct questions to ask.

Then going a step further, as soon as we have our M&E plan, we have also a data model. Why is a data model important? It gives us the information flows from the data collection to the data use. It provides us also with the information of how different data collection forms are associated. And this enables us to have consistency, to streamline basically the data collection. And this gives us consistency across data sources and also higher data quality.

These two components are the main sources of information or the basis in order to be able to ensure effective data visualization. Let's take a very simple and very common example. Example given: an M&E plan has an indicator which is "number of registered participants" or "number of beneficiaries." Let's assume now that this M&E plan states that we collect that indicator daily, we choose to analyze monthly, and we have two different needs. The program team needs the monthly monitoring disaggregation per partner for their own purposes to understand what's going on in the project. The donor needs the quarterly calculation.

We have also a data model. So what does a data model do in this case? Imagine that now you have in that data model a data collection form where you register the participants with all the information. Having a data model, and broadly speaking an information management system in place, enables us to avoid double counting. This is one of the most important and one of the most common problems in the sector. We have a specific data source. So all of us know that when we want to see the beneficiaries, whether we work in the office or in the HQ, we're going to choose a specific data collection form and at the end of the day we structure our data into a usable format. We can use it within the system, or we can extract it to connect it to other systems depending on the data flows that we have established.

In a scenario like this, we will end up with two different examples of charts. The line chart in that case might show the monthly variation across the two partners throughout the year. The donor report, because at the end of the day the quarters are a definite number of choices and options, can easily convey the message accordingly.

00:11:39 Principles of good data visualization

Now let's go and move into principles of good data visualization. As I mentioned, what is our starting point? The audience. Because at the end of the day, they're going to read each dashboard, each report. So what is important, and you need to have in your minds, is that you have different teams working internal and external. You have different people with different perceptions, different cultures, and different capacities for reading and understanding graphs. So the most crucial component is actually to match the visualization to the needs and understanding of an audience.

Example given: you have a report, as I mentioned earlier, that the field supervisors read monthly. You don't expect, neither is it needed for them, to have advanced knowledge for interpretation. So what is needed there is a very simple chart that can convey the message and possibly a text that showcases quickly that, look, there is a variation this month, or there is an increase or a decrease or a difference between your two partners. Imagine in the other case that you have a report that actually targets your donor, aiming to illustrate this quarterly progress. So as a consequence, you match the specific graph and the extent to which you add extra description and extra information to the purpose of who is reading that report.

Understanding the target audiences and purpose is actually crucial. The starting point for that purpose is the monitoring and evaluation plan. We frequently ask ourselves: which are the stakeholders that need to have timely information? When do they need to have information? Who are the stakeholders? What is the level of understanding? Do I need different reports? Can I create only one report for the field supervisors and the donor or not? Probably "no" is the answer. I need to separate reports based on why I'm designing the report. Is it progress to donor? Is it progress to HQ? Is it monitoring because supervisors need to be aware of what's going on in the field? All these questions help us understand a dashboard, the data visualization, and choose the right type of chart that matches the needs of the specific audience.

Example: I clearly had an intention here to overcomplicate that specific question. Imagine I put it into the extreme that we have a very common question, especially for cash distribution projects: "How many beneficiaries have their basic need met as a result of a cash distribution project?" Here I chose on purpose a very complex graph. If you ask me even to read it quickly, I cannot do it in less than a minute. The primary objective is actually to choose that specific chart that helps convey the message as fast as possible. So never complicate it. Less is more in the case of data visualization.

Another example, and this was thanks to UNOCHA working in Sudan and their partners for building a very nice publicly available dashboard on multipurpose cash assistance. Here we can see very clearly what is of interest: how many people reached. We have a nice pie/donut chart that shows us what is the percentage of each reporting organization. We have the numbers, the figures that clearly illustrate how many states, how many localities. This is an example of a very clear dashboard that conveys the message very easily. This is what we want at the end of the day.

00:15:42 Choosing the right chart

Choosing the right graph starts with the question: what is the main question that we want to answer? We always start with that question. Remember I mentioned the M&E plan. In reality, your question frequently, not always but quite frequently, coincides with that indicator. Number of beneficiaries, number of beneficiaries disaggregated per governorate, percentage of people that had their needs met within one month of cash distribution, disaggregation per partner, disaggregation per country. All this information sits within the M&E plan. And this dictates the main question.

From the main question, we understand the relationship that we want to have. What do we want to do? Compare between partners? Track over time across months or throughout the years? Correlate? See whether one intervention has a positive impact on X, for example, an increase in earnings? See how my data distribute? Compare a subset of data to a whole amount? Examine deviation? Rank a variable? So the main question leads automatically to the relationship.

And then on top of the relationship, we have a data type. So what is my data? Is it data that can be counted, can be measured? Is it a range of values? Is it a finite number of options? Can I group the data per category? So what you need to have in your mind here is: think of the main question. If I combine the relationship that this question gives you implicitly with a data type, I can safely proceed in choosing the most appropriate visual.

What are the most common chart types?

Pie charts: They work very, very well with proportions, especially when our categories sum up to the whole amount (100%). Example given: age group. I have separated my population into age groups: 18 to 25, 26 to 55, 56 and over. There, the pie chart and its variation, the donut chart, can perform pretty well. The main question here is: what proportion of the total does each category represent?

Bar charts: Here it's questions related to how different categories can compare in terms of a specific value. Example given: I want to compare and see across the months. Bar charts work pretty well with up to 50 categories. If we want to see across the months what is the difference of beneficiaries, a bar chart is a very good way actually to visualize that information.

Stacked bar charts: From a bar chart where you look by value across one category (basically months or different partners), in a stacked bar chart the difference is that you add more categories. In reality, now you can compare across 2 or 3 different categories across a specific total value. Example given: monthly beneficiaries by project. Let's assume that we have three projects and in all these projects we have a common target population. They perform different kind of services. And we want to compare across the months and to see how much each project basically contributes to this total of beneficiaries per month. Again, the stacked bar has two different variations: one that is the raw number and one that sums up to 100%. Both of them can illustrate very well how much each subcategory contributes to each total. It's a matter of the objective of why I'm designing it.

Line plot: It's great if you want to understand how data changes over a continuous period, especially over time. A line chart may illustrate different trends, patterns, fluctuations. Example given: let's assume that we run a project over multiple years, say five or ten years. How has the beneficiary number changed over those years and across all my partners? Imagine that here on top of that, this line plot for each specific line illustrates specific partners that we have worked with throughout the years. So it helps us understand what is the trend, how has that number of registered beneficiaries changed.

Histogram: Those are a great choice when we are asking about distribution or frequency of numerical data. Now at first, histograms can look like bar charts, but we need to have in our minds that each chart serves a different purpose. Histograms display numeric data grouped into continuous bins, which is useful when we want to see the distribution of values. In contrast, bar charts are used for categorical data where each bar represents a separate category such as different regions, the months, or different partners. Avoid mixing them. Try to make the differentiation between the two because if we misuse the two different subtypes, it can lead to misleading messages. Example given: wages distribution. Let's assume that we were working in a project and we have registered participants. As an outcome based on the service delivery, those participants will increase earnings or find a job. We want to see what was the wage distribution. In that case, we create different bins. Imagine one participant will say I receive 10 euros, the other 20, the other 100, 150, and so on. So it helps us put establish small bins. Let's say I establish the first bin from 0 to €100, the second bin from 100 to 200, and it's easier to analyze and see what is the distribution of that wages.

Scatter plot: Scatter plot is a very good choice when we have a question about two numeric values and how the specific variables associate to one another. How does a numeric variable A relate to numeric variable B? In this case, we have an example of a wage gap. What was the correlation between the wage gap in 2000 to the gender gap in 2015? The best practice is always to fit the trend line in order to quickly understand if the correlation is positive or negative or if there is no correlation at all.

Maps: We have very different maps, tons of maps. Very commonly, what I've spotted is that the bubble map is a very common scenario. Here we want to see how geographical information, how the variation of a numeric value varies across different regions. Example given: we want to see what is the incident of violation per province in a specific country in Africa. The bubble quickly shows us where we need to draw our attention, in which region.

00:25:37 Best practices for clarity and consistency

In terms of clarity and consistency, what you need to have in your minds is that the most important thing is actually to keep it clear, keep it simple. We need to carefully select only the data that is useful in order to convey the message. Otherwise, if we find ourselves in a situation that we overburden a specific dashboard, a specific page, a specific report, we run the risk that our message will not be delivered at the end of the day.

Again, I use this very nice dashboard they made in Sudan. Here we have a lot of information indeed, but very clearly structured so that we can see the different people and how the people vary according to each state or its organization and how they distribute within Sudan. Also, what were the people reached and the amount that they were dispersed across all the months. So clarity and consistency actually enable us to convey the message by adjusting the meaning to what we do.

A color always has a meaning. It's not randomly selected. It enables inclusivity also. So imagine cases with people that have color blindness. Of course, if you know that your target audience has color blindness, then there the color plays indeed the role because you want to help them interpret a specific graph. Also, the color and the style is a matter of culture. So we need also to be conscious and respect each kind of idiosyncrasy depending on where we are working.

Let's see the color first. Very important is to have sufficient contrast and separation between elements. We want to be sure that people can read whatever text is there irrespective of the color background behind. As I mentioned, color conveys meaning also. And that is why it enables inclusivity. In the first scenario, we have categorical data that do not have an inherent ordering. So each color means something different practically and there's no association across them. And it's easy to understand because indeed we have picked very different colors. The second way, the data is numeric or they have a natural ordering. So you see, it goes from very light to very dark. So it helps us understand that here indeed we have an ordering. In the third graph, data diverge from a certain value. And that is why we have chosen a different color. The right on the left also uses color lines. Imagine that case that we have the number of beneficiaries across partners. So each line at the end of the day can illustrate a different partner. And it's very nice, I would say it facilitates the reader if we use colors for specific categories.

Also, color is a cultural topic as well. And one example that I found, and I quite like it, is that pink color. We use that in the West; the pink is feminine, with a girl, with a female. And I was reading through some nice documents that I found online and actually they said in Japan it is equally used between masculine and feminine. Which was very interesting because it made me think of all the dashboards that we made with blue and pink for boys and girls.

Then labels and descriptions. Each data point needs to have a call out of the amount and the category. These are the most important things I would say: category and amount. Because we want to actually support the reader to help them start very easily, timely, without spending too much time. We need to have clear text that labels the significant parts of the data. Imagine given in the second chart, we have the beneficiaries presenter and per activity. We clearly state which color is male, which color is female in order to make the points. And of course, we always state what are the Y and the X axis.

Font size is important. As the rule of thumb, I would say use 12. But keep in your minds that whatever you make, you need to check it that people can read quite easily. Going one step further in the labels and descriptions, what I have found a lot useful is actually the use of alt text in order to describe different kind of information that you find in the map or different conclusions that you have for a specific chart. Chart descriptions also can serve the same way. So depending on the software that you will use, at the end of the day, you may have cases that you will use an information button or simply you will write a label, a text with more information for a graph. White divider also I have found it quite useful. It helps differentiate especially in bar charts so people can read more easily.

00:31:59 Identifying good versus bad visualizations

Let's see some examples of good and bad visualizations to have in your minds.

00:34:04 Data visualization examples

There are two different and very commonly used scenarios in information management systems when especially we want to address the whole data circle from data collection to data use.

The first scenario says that an organization frequently finds themselves using a platform for the data collection, ActivityInfo for example, plus another system for data visualization like Power BI. Here it's a very common scenario. We need to have in our minds that always integration is needed. So as soon as we have two systems in place, we need to maintain also two systems in place. This implies that on one side, a dedicated system such as Power BI offers N+1 different options for data visualization. The other thing that we need to consider is that we need more time spent in integration. We have a higher level of capacity building internally because we need to capacitate people in both softwares, and higher resource allocation basically.

The second scenario, and also a commonly used scenario, is having a data collection system and data visualization in the same place. ActivityInfo, as also Brendan mentioned in his introduction, can be used for such a case. So in that case, no integration is needed. And this implies in turn that we need to spend less time, we need less resources, less budget, and less capacity building at the end of the day.

Having said that, I want to give you three different examples. The first example showcases a development assistant project. The second, a cash-based intervention project. The third is a bit more generic, more cross-cutting: feedback complaint response mechanism.

Example 1: Development Assistance Project This specific template was designed as an example for a program for promoting Pennsylvania Maple Syrup. Here, imagine in order to provide a bit of context that we have few indicators such as number of farms enrolled, number of individuals benefiting or accessing agriculture-related financing. And we track a few indicators also regarding the outcomes, like total yield per farm increases 20% among participants. Here the added value of having everything in one system is that we can use the information very timely. And here we have a specific dashboard that shows different distribution location of farms in a map, the bar chart. Then we have a next page that showcases the trend line of different users disaggregation.

Example 2: Cash-Based Intervention I think that they have always a strong preference for the cash-based intervention for some reason; it's a very common project in the humanitarian development sector. In that case, we track different kind of indicators: number of people registered, receiving cards, etc. For the purpose of that specific template, we had created two different reports. A monthly monitoring report that includes different kind of tables combined with different kind of graphs, like the common pie chart. And the post-distribution monitoring report that also helps us to add notes, as I mentioned, the details of a graph. But in a dashboard, use a combination between different kind of labels, tables, and charts at the end of the day.

Example 3: Feedback Complaint Response Mechanism This is a more cross-cutting situation. The feedback complaint response mechanism has to do with the policies within our organization. So many organizations collect feedback from the beneficiaries and they need to respond back timely to this feedback. And frequently you have the sensitive and non-sensitive categorization. Here we have also different data collection forms and an internal report that showcases the different channels, the different categories, the count of the different services or different indicators or different surveys that we collect frequently after we have delivered service, after we have addressed the feedback.

00:42:05 Key takeaways

I think that the most important for me is to keep 4 different key messages for today's presentation.

At the end of the day, the most important thing for me is always to consider the data visualization from the data collection. In reality, the planning for how I'm going to use the data starts with that first time that I collect the data or even before collecting the data. It starts at the time that I'm planning for that data collection.

00:43:47 Q&A session

Is it professional to include more than five visuals in a report? What is the recommended number? I don't think that there is a recommended number of visuals within the report. I think that has to do with the specific format and specific page space that you work within. So including information is a matter of balancing out how hectic that page can be and how much information you can include. It's better to include what you need. So if you go back and you ask what is needed, then there is information for the starting point. Some software maybe would allow you to change the page orientation to include different kind of graphs. I would say like frequently I have found myself to combine no more than two big visuals in the same page but with addition of text, with a number, with something that provides more information. But this is kind of a rule of thumb. For me, the important thing is that if you respect the spaces, then this kind of rule also plays a filter to how many visuals you include in a specific page.

Is it professional to add data values to bar chart or stack bar charts? Yes, definitely adding data values is recommended. For the sake of my presentation, I simplified the graphs because when I was using the stacked bar I wanted to place your attention not on the formatting on that specific slide, rather on the essence of the stacked bar. But generally yes, data values should be included in the charts.

Which chart is best to visualize advocacy data? This is very generic but it depends on the question. If it is attitude change among stakeholders, I cannot answer because attitude change depends on how you measure attitude. So imagine I can choose a chart as long as I have a question and I know the data type. So in that case, I miss the data type. I'm not quite sure how attitude is being measured. Is it one question? Is it an index? What is that attitude? So in that case, it's hard for me actually to provide a specific answer. But if you can provide me more information on the data type I can be more specific.

What is the best analysis tool for a data set around 500 entities? I would say ActivityInfo is an ideal tool for 500 entities. Generally speaking, 500 entities is neither a small amount but neither a huge amount. In that case, it lies somewhere in between. I mean on my side I could recommend ActivityInfo for the start from data collection to data analysis. And I think that it will be sufficient for those 500 entities.

Which software was used to create the charts and graphs being presented? The graphs and charts, because I wanted to showcase and have exactly the same charts and graphs with specific colors, and I didn't base my analysis on any data in that case, we used picture-related processes because exactly it was for the purpose of the specific webinar. So I didn't have any data behind. Now for the ones that I showcased on the template, it was ActivityInfo. For the ones that I showcased the screenshot from Sudan, it was Power BI.

Which is the best way to present the data percentage for one category as a total? It could be depending on how many categories you have. If I want to present percentage of one category I will choose a donut or a pie chart depending on the number of categories. If it is percentage with that category along with male and female, a bar chart would be used here in that case, or a stacked bar chart better because you have male/female and you have different categories. So the best one would be the stacked bar chart.

Does one need extensive training in programming to be able to use ActivityInfo? The short answer is no. The extended answer is that you need some kind of familiarity. You need to be a bit familiar with the use of the system as every new system, but programming skills? Definitely no. Neither extensive nor basic I would say.

How are you going to share the link of visualization page with others? Different kind of dashboards, and this applies also in ActivityInfo, can be stayed within the specific information management system AKA database for specific users that have access and are system users. Or you claim those dashboards or reports publicly available because they don't include personal identifiable information. They include different kind of graphs and you generate a link via which you can share or you can embed in a specific page or another report or you can share with someone else that is needed.

How do you strike a balance between decluttering and showing data labels for data polish? I think the balance comes from the fact that you need to choose the right graph. Meaning if I was choosing a pie chart with 10 categories that will be chaotic in the first place. And there you had a problem also with the labels. That is why the pie chart has a maximum of a few categories, 5 or 6 maximum. Similar with the bar chart. So we strike a balance before because at the end of the day each graph is designed for specific purposes. The starting point is to choose the correct graph. If you choose the correct graph then I think that you can strike a balance between the data labels for all the data points within the chart.

How do we count or measure indirect beneficiaries for coastal communities? We don't have anything specific for coastal communities to the best of my knowledge, or if you refer to a specific project. But in that case, the templates serve as an example. And the templates can be customized for any other case or any other target audience. So everything is customizable to the needs of a specific project in place.

Sometimes we need to combine complex and simple data in one visualization. How can we make it easy to understand? Usually, data visualization will adhere to one data type. So if you try to combine it, clearly the result will be very misleading. That's not the best practice. Broadly speaking, avoid combining. You have a graph actually it's there for specific data types.

Which tool do you use for visualization on maps? ActivityInfo has data visualization on maps that I think it's worth checking. It has some very, very nice options.

I'm asking about qualitative data analysis and visualization. What system are you using? ActivityInfo primarily is with quantitative data analysis and visualization. I would say that I'm not an expert in qualitative, but to the extent that I have used it, there are different charts and graphs that perform the word count and it showcases what is the most common word that you can use. That's the most frequent chart that I would use in that case.

Should we write the number of chosen options for multi-choice questions in percentage or number? Depends on the objective. What is your question? Do you want to know what is the percentage of X to the total or the raw number to compare? So the choice between whether you go with a percent or number depends on what is the question that you pose.

Should we use word cloud for multiple choices question or bar charts? Depends also on how many multiple choice questions you have. So if you have a few, a definite number of multiple choice questions, a bar chart can work. If you have an excessive amount of numbers, like 20 or 25, then there the word cloud would convey easily the message.

In the choice of visualization software, to what extent does cost and fund sustainability come into play? I would say that when you go with two softwares, in terms of cost you need always to consider double the license. That's the first and foremost cost. And then you have different kind of indirect costs: the time needed to train the different people on the two softwares, the time that the staff is going to spend on the maintenance of both softwares, the time that staff is going to spend on the integration between the two softwares. So there are some direct licenses and costs of the two softwares of course and some indirect costs to this center.

Is there any easy software used to make that graph without much operations? I'm not quite sure what you mean. I would say that always you need to put some effort. The effort is minimum on the choice of the correct chart as long as the data are cleaned in reality. So the effort is not on the data visualization; that is the fun part. The effort is usually before the visualization, the data collection, especially the data cleaning in the absence of a correct tool.