Triangulating data

When it comes to my core beliefs about data, the idea of triangulation is one of my non-negotiables. Triangulation originally started in military and navigation contexts, as it was used when people were navigating with a map and a compass. If you were lost, as long as you could see and identify three landmarks that were on your map, you were able to use the map and compass to ascertain your position.

If you could see for example, the peak of a hill, you could use your compass, measure your bearing to that hill, and draw a back bearing (straight line) onto your map. That data point was useful because it would give you some indication of your position, but the problem is that the angle to the landmark would be the same if you were a hundred metres away from the landmark, or a thousand metres away from the landmark. Therefore, that single data point alone was not hugely useful because it wouldn't tell you much.

You would then take a second bearing to another landmark. If you could see a power pole that was marked on the map, you could again, work out your bearing to the power pole, calculate the back bearing, and then draw the back bearing onto the map. The map would now have two straight lines, and ideally, they would cross over. You might think that the place where the lines cross is your position, and that might be the case. However, if you've made a mistake with either of those bearings, what the lines are telling you could be very, very wrong.

By taking a bearing from a third landmark (for example, a lighthouse), and drawing the back bearing on the map, means that you now have three straight lines, and a triangular intersection on the map. The feedback to the user at the time was that if that triangle was quite small, the three readings were quite accurate and the given position was accurate. But if the triangle was large, it indicated that (at least) one of the bearings was wrong.

At the time, the navigator would have had to reestablish the bearings and the back bearings, check the lines on the map and then hopefully find their mistake; if so, they could correct the issue, draw the correct lines, and the triangle would shrink.

Thankfully, we don't use that technique much anymore when we're navigating, as our phones, Google Maps and GPS on our phones do the heavily lifting for us. But we do still see triangulation used really heavily in research. In research, outcomes and recommendations are never based on just one trial, one person's experience or what's happened, or on one team or one organisation. They're always a collection of multiple perspectives, often from multiple sites, maybe multiple divisions within the site, or multiple data sources, that are drawn together to generate the findings. Researchers look for similarities and differences across a range of perspectives and data types. It doesn't really matter how researchers triangulate (whether by group, site, or data source), but it means that the research findings are more valid and reliable, and more replicable to other contexts.

In our work, and just like in our physical map, it is important to think about the types of data to use in triangulation, that will help broaden our understanding of the phenomenon that we're investigating. The data sets will be different (of course), but we want the three to align as closely as possible. A good way to think about it is, 'what three sets of data will tell me the most useful information about [this focus]'?

It's important to point out that that data sources used in triangulation don't have to be quantitative; they could be qualitative information. When we triangulate, we choose three data sources that are quite similar and that are interested in the same thing and focused on the same area of inquiry. It might be that some qualitative data does that best. (The caveat here is that the use of, and engagement with qualitative data can be a lot harder and more time consuming than quantitative. It's important to consider the cost-benefit ratio of including this type of information in this process.)

When we have our three sets of information, we start to think about, and look for trends and insights in the data. It might be obvious that across those three data sources, you're seeing the same things, and the same messages are across all of them. If that's the case, that's brilliant, but it doesn't always happen.

It is more likely that you will see a trend or insight appearing in two out of three data sets, or in one set only. If that's the case, that's actually okay - that's why we are triangulating! In this case, we want to trust the MAJORITY of the data. We trust the two pieces of data or the two data sets that are telling us a particular thing, and we can discount the third. If you are using more data sets or points, the same is true - if more than 50% of your data is telling you a particular thing... Trust it!

One of the challenges people face when they are new-ish to using data is that they often worry about missing something. They worry about looking at a data set and not seeing something that they should have, and then investing time and effort in something that's incorrect, or ineffective... While this is a valid concern and something to be mindful of, triangulation is a way of alleviating some of that concern, as we don't just focus on one metric, but look more broadly across a series of data sets.

In a perfect world, and if you have a relatively small team or relatively small data sets, I like the idea of having data in the same spreadsheets, sitting in the same heat-mapped table, side by side. If you're looking at say, data for particular retail stores, or if you're looking at different metrics that apply to different employees, being able to see the triangulated data side by side is a really important part of being able to use the data well. What we DON'T want to do is have to jump back and forth between screens or between different visualisations to work out what is happening in the first set, what is happening over in the second, and then what is happening in the third. It makes the process of triangulation and looking for trends and insights much harder.

An example that I often go to on the idea of triangulation relates to our personal financial situation. If I asked you about your financial position right now, but all you did was told me how much money was in your transaction account, it may not be a very good indication of where you're at. You might have quite a good amount of money sitting in your transaction account, but you might have a huge mortgage that you're behind in, significant credit card debt, and/or car loans or student loans... Or equally, you might not have much money in your transaction account, but you've invested well, you've got some money invested in shares, and you have a very small mortgage as you've been paying it down for years.

It's a good example that no single metric is particularly useful in reflecting your financial position. However, by bringing in a couple of different elements and pieces of data, we can get a much more well-rounded picture of our financial position.

So, when you are investigating a particular element of inquiry or a particular phenomenon in your work, think about three (or maybe four or five - but that's enough :)) sets of data that tell you the most useful information about the thing you're investigating. Use the language of triangulation with your team, with other people, and help them get comfortable and confident with drawing on a range of sources of information and then drawing out of those sources collectively, the trends and insights. We need to get this initial step right, so we make more effective decisions about what to actually do with the data.