Chapter 17 FAQ

What is COINr?

COINr is an R package for building and analysing composite indicators. It aims to make many (sometimes complex) operations in building and auditing composite indicators fast and easy to reproduce.

What is a composite indicator?

A composite indicator is an aggregation of indicators which aims to measure a particular concept. Composite indicators are typically used to measure complex and multidimensional concepts which are difficult to define, and cannot be measured directly. Examples include innovation, human development, environmental performance, and so on. Composite indicators are closely related to scoreboards, which are also groups of indicators aiming to capture a concept. However, scoreboards do not aggregate indicator values. Composite indicators also usually use a hierarchical structure which breaks the concept down into elements, sometimes known as sub-pillars, pillars, sub-indexes, dimensions, and so on.

If you want to know more about composite indicators, see the European Commission’s Competence Centre on Composite Indicators and Scoreboards and in particular the Handbook which can be found here with a few other useful things.

Do I need to be “good at R” to use COINr?

COINr is designed to have a simple interface which mostly involves passing COINs from one function to another. This means that you can build a composite indicator in a few simple commands. You can also export easily to Excel and prepare graphs and tables there if you prefer. Of course, composite indicator construction often involves iterative steps and it is likely that some things will have to be done outside of COINr. So the answer is no, but it helps to have a basic level of competency in R. See Appendix: R Resources for some suggested ways to get more into R.

Thanks but is there an equivalent tool in Excel?

Yes - the COIN Tool is an Excel tool which does similar things to COINr, although the options are more limited. It is however a very useful tool and if you prefer to do everything in Excel this should do the job.

What are the steps for building a composite indicator in COINr?

See Appendix: Building a Composite Indicator Example.

What are the steps for analysing a composite indicator in COINr?

See Appendix: Analysing a Composite Indicator Example.

I am having trouble assembling the COIN. What should I do?

First, make sure you read the COINs: the currency of COINr chapter. Next, I would really recommend looking at the example data sets, ASEMIndData, ASEMIndMeta, and ASEMAggMeta. If it is easier, export these to Excel (using e.g. write.csv()) to look at them there and perhaps use them as a template for your own index and data. If you are having problems, it may be in the definition of the index structure. Call ASEM <- build_ASEM() which builds the example COIN, and then plotframework(ASEM) to see the structure of the example. Now look at the ASEMIndMeta and ASEMAggMeta data frames. Have a cup of coffee or tea and mull it over. Hopefully it should make sense!

Can I construct a composite indicator over multiple years?

At the moment, COINr does not have comprehensive support for panel data, i.e. data that has different values for different years. This can also be complicated because from one year to another, indicators may change, there may be a different set of units, methodology may change, and so on. A suggested workaround, for the moment, is to build a separate COIN for each year of data. This allows methodological variations if needed, and the results can be compared using COINr’s compTable() and compTableMulti() functions. In recent versions of COINr, the assemble() function gives the possibility to extract and impute single years of data from a panel data set. See COINs: the currency of COINr for more information on this.

You would probably also need to extract results using your own code as well.

If you follow this approach, it is a good idea to store COINs in a list (or possibly a data frame), which means you can apply COINr functions to the whole set of COINs using e.g. lapply() and purrr commands. Eventually, time permitting, I would build this into a package. See the Many Models chapter in the R For Data Science book for a bit more general information along these lines.

Do I have to aggregate indicators? Can I just make a scoreboard or only aggregate up to a certain level?

No of course, you don’t have to aggregate at all, and in some cases it might not even be a good idea. However, you can always see what happens when you aggregate. You can also aggregate up to an index, but choose to only present pillar or dimension scores. This might make sense, for instance, when pillars represent very different things and are poorly correlated or divergent between one another. Another thing to keep in mind is that an aggregated score and a scoreboard are not mutually exclusive - you can easily present the two together.

I found a bug - what do?

You can open an issue on the COINr repo. Otherwise, you can also just email me.

I need help. Who ask?

Please search this manual first and also the help documentation by typing e.g. ?normalise (depending on the function you are interested in). If you need general assistance with composite indicators and indicators, email me and I may be able to help.

Composite indicators are full of subjective choices, so doesn’t that make them pseudo-science?

Composite indicators are indeed full of subjective choices, and this introduces uncertainties in their results (scores and rankings). For example, weights, aggregation methods, which indicators to include, are some of the many choices to make along the way. However, if you have worked in any modelling exercise, you will know that any model has similar problems – which processes to include, which modelling approach to adopt, what values to assign to parameters, and so on.

This means that, like any modelling exercise, composite indicators should be interpreted carefully. Which is to say that you should be aware that small differences in scores and ranks may not be significant. Uncertainty analysis can help with this.

It’s also worth pointing out that composite indicators usually aim to measure things that cannot really be measured any other way, like innovation or quality of life. And also that even if we build an index, it should always be presented alongside its underlying data.

If you want to read in more detail on this, you could see my blog post on the topic.

COINr is great and I want to cite it. How do I do that?

At the moment there is no official publication for COINr. There is this online book, but it is not “officially” published. A paper on COINr is currently (Oct 2021) under review in the Journal of Statistical Software. If it makes it through, I will put the citation details here. In the meantime, follow something like this.

I get errors when installing COINr.

The CRAN version of COINr has survived many CRAN checks, so it should install. If you get errors, it is most likely a problem with your R installation rather than a problem with COINr. I have also experienced problems installing packages sometimes (probably everyone has).

First of all, you may be asked to update some packages which COINr depends on. This is a good idea, but it could cause problems. If you update many packages at the same time, things seem to have a habit of going wrong. I have found that it is often less problematic to update packages individually. Sometimes, even individual package updates may have problems due to packages being “locked” or it not being possible to write to package files. In this case, the best thing seems to be to close R, manually delete the package from your R Library folder (i.e. delete the package folder), reopen R and then install it afresh.

It will also often help to update R and R Studio to the latest versions.