Key to produce credible estimates but only worth it if we communicate them properly

Graphs and tables are everywhere in economics

They are often the result of analyses

Readers often start looking at graphs and tables

Seems pretty simple, we all know how to make graphs and tables

Sure BUT handled carelessly a graph can be inefficient and even misleading

Once you pay attention to data viz, it is fun, instructive and satisfying!

I will probably state obvious stuff BUT

• Maybe still good to hear it sometimes
• I will provide references to explore all this in more details
• I will try to structure and conceptualize

Objectives of the class:

• Provide tools to nail your data viz and your tables
• Communicate that how you present results matters and provide a conceptual lens

Why good data viz matters?

What are the characteristics of good data viz?

How to make an ok data viz?

Communicating in economics

Main take-away points

Allows to synthesize large amounts of information into a limited space (that also applies to tables)

Ease interpretation (reduce the cognitive load)

For your audience to not only receive info but also understand and remember it

Graphs for you (and your future self): can be quite rough on the edges, but you will want to be able to understand it in the future

For presentations: you have some leeway for explaining orally your graphs

For the paper: there is only a couple of graphs in a paper; make them perfect!

Beyond credibility, why make nice looking visualizations?

To trigger interest, to intrigue, to catch the eye

That affects how people perceive information

Nice looking visuals may be more memorable

They can help engage the audience

Pretty does not always mean non-simple. Simple graphs have value.

In academia, maybe better to keep the design rather minimalist

With the same data, you can:

• Tell a lot of different stories

• Emphasize different points

Making a graph = choosing a lighting for your data

The type of graph you choose matters and depends on your story

Intrinsic load: inherent complexity of the information

• eg, 3/7 more difficult to compute than 3 + 7

Germane load: how familiar is the audience with the new information

• eg, box plots, coefficient plots might be difficult to understand without prior experience

Extraneous load: how the new information is presented

• eg, horizontal text, highlight important data points, annotate, etc

There is actually a lot of theory behind data viz:

• Perception,
• Colors,
• Design,
• etc

Can relate to cognitive load

But can also leverage it to make better data viz

Worth learning about it and being aware of key principles

Explain orally what your graph represents!

• What is on the x-axis?

• What is on the y-axis?

• What is the message you want to convey?

Know what is possible to do to find what you need

Know graph names (to search the internet regarding how to code them)

Refer to existing graph (type) galleries:

• Data Viz Project

• Flowing Data - Chart type

• Dataviz Inspiration

• R graph gallery

If you start paying attention to data viz when you see them, you will see what works, what does not

It is a process that takes place in the “background”: you will learn quickly and not realize it

You will see nice looking graphs (and hopefully enjoy it)

You will build better and more impactful graphs

Original graphs may trigger interest

Familiar graphs may convey the point more easily

My take:

• Use the best type of graph, regardless of its originality/familiarity

• If it is different from what people are used to, make it easy to read

Often have a specific type of output: estimation results

There are norms for tables, in particular: deviating from them is costly

And a massive number of observations

Our analyses rest on identifying assumptions \(\to\) can explore with graphs and tables

We often use very complex models. Visualization can help understand what we are actually estimating

As rhetorical visualization tools for models

To explore our data

To check the validity of our models and assumptions

As diagnostics

To communicate results

We will run a simple analysis:

• Link between jail sentences for drunk driving and traffic deaths?

At each step, we ask the same two questions:

• Which graph helps here?
• Which table helps here?

Side goal: provide practical tips to make these tables and graphs (eg chunks of code, packages, etc)