Can you make a data application as good as FRED?

What makes the Federal Reserve’s data explorer so good?

FRED is the St. Louis Fed’s tool to explore economic data series, manipulate them, and then visualise them. It makes it easy to generate graphs like this:

Relative inflation of apparel vs all goods in the US

FRED is an undeniable joy to use. It’s really fun to explore these datasets. What makes the FRED experience so good, and how can we learn from it?

The interface makes it easy to perform the operations you likely want to

a. The interface follows well established metaphors to deliver the functionality

b. Accordingly, it is fairly easy to explore all the options available to the user

c. Use of algebraic input to manipulate the data series (normalise, combine, etc) means users are using a metaphor they know very well. Attempts to make this easier generally make it harder for everybody.

d. Almost everything has contextual help, which provides enough description such that the user can start anywhere and gather context about the whole application.

Takeaway for us

Try to use an off the shelf interface that has documentation and a strongly consistent interface. If you choose to make your own, less is more in many ways. Taking that into account, don’t be afraid to create your own custom, lean interface.

The datasets are richly indexed, with tags, date ranges, sources, and descriptive names. Most datasets also have textual descriptions about their origin.

Unfortunately most organisations don’t understand the benefit of data cataloguing and curation. The ability to easily discover data sets that one wants to work with is really essential to FRED appealing to anyone whose job isn’t using FRED every day. If you don’t believe me, compare the UK’s equivalent with FRED.

The ONS Time Series Tool seems similar, but just isn’t as easy. The selection area on the left is static, rather than dynamic. In FRED it changes dynamically to reveal more detailed tags. The selection tool in FRED shows you all your options (at the current level) all at once; the ONS tool just has drop downs, so you can’t look at the options except by interacting with the drop down specifically. No leisurely consideration for UK researchers.

The FRED display of each search result is taller — it takes up more screen height. The big letters make it very easy to scan the titles. The titles are pretty good, although some are cryptic. You always get the date range, and you get a description of data frequency and units. The ONS results probably contain more information to those well versed in the terminology used. However, these are all in codes. You won’t see the date range available, but you will see the publication date.

Finally, units — Fred knows what units a series is in by default, and how to transform it to other types of units.

Takeaway for us

Establishing a standard to describe the data is important; but the content of the standard does matter. Consider who your users are, and what they really need to see. The ONS tool is clearly designed for people who use it every day. FRED is clearly intended for the professional user, but accomodates new users.

Consider making data curation a part of your data users job — think of something like a wiki, but structured, and integrated with search and data visualisation experience.

Unfortunately there are not good open source data cataloguing tools, and even commercial BI tools have very uneven built in cataloguing. Looker is built around data curation, but there is really no way to publish the metadata to ones users. Domo has pretty good support here, but isn’t super smooth.

There are commercial tools in this space. You’ll want to carefully evaluate the experience of both consuming and publishing data through these platforms.

Almost nothing in the way of bells and whistles beyond the core functionality

FRED lets you do three things: find data sets, manipulate and visualise them, and download or publish the visualisation. Other than displaying some metadata and customising the display, that’s it.

What’s good about this minimalism is that the less you can do, the easier it is to figure out the functionality that is there. There’s only so much clicking around you can do before you learn enough about the interface to get oriented.

TL;DR

Keep your data applications simple (small, coherent set of features) and make it easy for users to find the data sets and keep them in good order for themselves and their colleagues.