Category Archives: Excel for Evaluation

Four Steps: Social Network Analysis by Twitter Hashtag with NodeXL [Guest post by Johanna Morariu]

Note from Ann: Today’s guest post is from Johanna Morariu, Director of Innovation Network, AEA DVRTIG Chair, and dataviz aficionado.

snaBasic social network analysis is something EVERYONE can do. So let’s try out one social network analysis tool, NodeXL, and take a peek at the Twitter hashtag #eval13.

Using NodeXL (a free Excel plug-in) I will demonstrate step-by-step how to do a basic social network analysis (SNA). SNA is a dataviz approach for data collection, analysis, and reporting. Networks are made up of nodes (often people or organizations) and edges (the relationships or exchanges between nodes). The set of nodes and edges that make up a network form the dataset for SNA. Like other types of data, there are quantitative metrics about networks, for example, the overall size and density of the network.

There are four basic steps to creating a social network map in NodeXL: get NodeXL, open NodeXL, import data, and visualize.

Do you want to explore the #eval13 social network data? Download it here.

Here’s where SNA gets fun—there is a lot of value in visually analyzing the network. Yes, your brain can provide incredible insight to the analysis process. In my evaluation consulting experience, the partners I have worked with have consistently benefited more from the exploratory, visual analysis they have benefited from reviewing the quantitative metrics. Sure, it is important to know things like how many people are in the network, how dense the relationships are, and other key stats. But for real-world applications, it is often more important to examine how pivotal players relate to each other relative to the overall goals they are trying to achieve.

So here’s your challenge—what do you learn from analyzing the #eval13 social network data? Share your visualizations and your findings!

Dataviz Challenge #6: Unit Charts

Lately I’ve been feeling let down by summary statistics: the min and max, mean and median, quartiles and standard deviation… They do their job well enough. Summary statistics tell a summary. An aggregate story, bringing all the messy scores together into some sort of cohesion. We grab the averages and stick them in bar charts.

But sometimes we don’t want to summarize, we want to highlight the variety in scores and remind readers that the chart is actually made up of individual people, not just the mean or median. Long live the messy data, the dispersion, the distribution, the spread!

unit_chart_1

I could tell you a few descriptive statistics: min = 26%, max = 100%, Q1 = 64%, Q3 = 83%, median = 74%, mean = 73%, standard deviation = 15%. Or, I could show you the spread in this unit-chart-turned-histogram.

Unit charts are not your new go-to chart. They do not replace bar charts. They are not appropriate for all datasets. They’re best for those few moments when you choose to emphasize individual units of data. A unit could be 1 person, or 10 people, or 1 school, and so on. Units can be represented in circles or squares or triangles. Units can be stacked on top of each other to form a histogram, or they can be plotted along a line.

The dataviz challenge: Re-create the chart in in Excel, R, or some other free software program. Then, tweet a screenshot to @annkemeryBonus: Make a unit chart for your own data. Or, do you emphasize individual differences with other chart types? Share your ideas with the community!

The prize for playing: A professional development opportunity and bragging rights. I’ll post the how-to guide in a couple weeks.

Want to learn more? I’m presenting about charting techniques at the American Evaluation Association’s annual conference on Thursday, October 17, 2013 at 11am in Washington, DC. Hope to see you there!

Dataviz Challenge #5: The Answers!

I’ve been in love with diverging stacked bar charts since I saw Joe Mako’s submission to Cole Nussbaumer’s dataviz challenge last December. Joe made this contest-winning chart. But in Tableau! The amazing but expensive software!

Could I ever create one in Excel?!

Yes! Luckily I’d learned about the Values in Reverse Order feature from Stephanie Evergreen. With Joe’s inspiration and Stephanie’s strategy, I started making these beauties for myself in Excel.

I wanted to share the chart secrets with all of you, so last month, I challenged readers to re-create a diverging stacked bar chart like this one:

diverging_before-after

It looks like I’m not the only one who loves diverging stacked bar charts. Congratulations to the 12 contestants! In order of submission, they are:

Most contestants seized the opportunity to use their own datasets and made adjustments as needed. For example, Sheila’s dataset fit a traditional stacked bar chart better than a diverging stacked bar chart, and Anjie needed to display cut-off scores.

So how do you make these diverging stacked bar charts, anyways?! There are at least two strategies: Either a) create two separate charts, a strategy demonstrated in previous posts like this one, or b) use floating bars, a strategy demonstrated in previous posts like this one. Stephanie Evergreen blogged about strategy B a few weeks ago and her explanation is pretty awesome, so I’m going to focus on strategy A today.

Here’s a slideshow about the two-charts-in-one strategy. Enjoy!

Bonus! Download my Excel file.

Want to learn more? I’ll be sharing my top 5 must-have chart strategies at the American Evaluation Association’s annual conference on Thursday, October 17.

For discussion: Nearly all of the contestants requested friendly feedback on their graphs. In most cases, contestants were trying these charts for the first time and thinking about whether or not these charts could be adapted for their datasets. What do you think?

Dataviz Challenge #5: Diverging Stacked Bar Chart

Last week I shared strategies for improving any chart’s colors. One of the examples was a diverging stacked bar chart:

diverging_before-afterI love stacked bar charts because they’re pretty versatile, and because they’re a great chart for lots of evaluation and survey data. In my example, I looked at the percentage of survey respondents who selected strongly agree, agree, disagree, and strongly disagree on a satisfaction survey. But stacked bar charts can be used in dozens of different ways.

So when can you use a stacked bar chart?

  • Stacked bar charts are for part-to-whole relationships. Use them when you want readers to see both a) one portion of the bar and b) compare that piece to the entire bar.
  • Stacked bar charts can be used for tallies or percentages. A tally is the number of actual people, dollars, etc. For example, a nonprofit could display their funding sources in a stacked bar chart – $100K from a foundation, $200K from a government grant, and so on. The reader can see the size of each grant as well as how the grants stack up as a whole.
  • Stacked bar charts can be used for nominal, ordinal, or diverging data. An example of nominal data is the racial/ethnic categories of your survey respondents. Ordinal data has a natural order – from best to worst, most to least, something to nothing – like my example. Diverging data is a subtype of ordinal data – when the categories are polar opposites and there’s a clear middle ground or neutral zone in between two ends.

And when can you use a diverging stacked bar chart? Diverging stacked bar charts are just for comparing several sets of ordinal data at once. They work best when you’ve got an even number of categories (like the 4 survey choices). Then, you can easily line up the midpoints along an invisible y-axis.

The dataviz challenge: Re-create the “after” version in Excel, R, or some other free software program. When you’re finished, email me or tweet a screenshot to @annkemery.

Bonus! 1) Adapt this chart for own data. Think outside the box! 2) There are at least two different ways to create diverging stacked bar charts in Excel. Can you find more than one solution? (And these charts are so awesome that you’ll even see one solution on Stephanie Evergreen’s blog next week!) 3) Don’t forget to use custom colors!

The prize for playing: Beer or coffee, my treat, the next time you’re in DC; a professional development opportunity; and bragging rights.

I’ll post the how-to guide in 3 weeks, on September 6. Happy charting!

Nominal, Sequential, or Diverging? Simple Strategies for Improving Any Chart’s Colors

Colors can make or break a chart. Colors direct our eye movements, and therefore our brains and attention. It’s up to you: will you help or hinder your reader’s understanding?

Here are some simple strategies for communicating clearly with chart color.

Strategy 1: Select a custom color palette.

Rather than using Excel’s default colors, match your chart to the organization’s logo. (Consultants: Match your client’s logo, not your own.) For my grad school projects, I align everything with my university’s logo.

Does the organization have a super basic color scheme? My grad school’s logo is green and yellow, which doesn’t give me many options to work with. So, I found a similar color palette on design-seeds.com. I used the instant eyedropper to find each color’s RGB code. Now I’ve got six colors to play with instead of just two.

chart_colors

Strategy 2: Figure out if your categories are nominal, sequential, or diverging.

Nominal or categorical variables are things like race/ethnicity (African American, Asian, Latino, White, etc.) or gender (male or female). Think about which pattern you want to emphasize, and use darker action colors to draw attention to that finding.

nominal_before-after

Sequential or ordinal categories have a natural order, like age ranges (5-9 year olds, 10-14 year olds, and 15-19 year olds) or years (Year 1, Year 2, and Year 3 of an evaluation). Sometimes categories go from less to more or from nothing to something. An example of a nothing to something progression is a satisfaction survey question that asks program participants to assess how likely they are to recommend the program to a friend (not at all likely, somewhat likely, very likely). For these charts, the action color can represent the something and white can represent the nothing:

sequential_before-after

Divergent categories are opposites, like agree/disagree scales on surveys. An example is a similar satisfaction survey question that asks participants to indicate whether they agree or disagree with the statement “I’d recommend this program to a friend.” When charting divergent variables, you might design a diverging stacked bar chart, as shown below. Select two different colors from your palette, like greens for agreement and yellows for disagreement. The extreme values (strongly agree and strongly disagree) get the darker colors.

diverging_before-after

(For a deeper discussion of these principles, check out colorbrewer2.org.)

Strategy 3: De-clutter by increasing white space and switching some black text to gray.

Which information is most and least important? Let’s de-clutter by removing or reducing anything without a crucial purpose. We want the reader’s attention focused on our most important patterns.

For example, if you’re using Excel, you might improve upon the default settings by deleting the border, the grid lines, or the tick marks. If you decide to keep the grid lines or tick marks, try adjusting them from black to gray so they fade into the background. You can also remove the legend and put labels within the chart itself (like that first bar chart with race/ethnicity information). Finally, you can outline shapes in white to give the chart a crisper look and feel (like the diverging stacked bar chart shown above).

Do you have additional color tricks to share? Please share your comments below.

Want to create one of these charts? Download my Excel file. Or, want a step-by-step tutorial? Stay tuned: Next week’s dataviz challenge is a diverging stacked bar chart!