New paper: a guide to representing variability and uncertainty in biodiversity indicators

---

Biodiversity indicators are metrics that show how a specific part of biodiversity is changing. Common examples include the Red List Index showing average species extinction risk and the Living Planet Index showing average change in vertebrate species abundance. These tools are used to inform important decisions about policy, resource allocation and conservation efforts. The key message of an indicator must be clear so we can easily see and interpret the state of nature. We can present indicators as a single line that changes over time as the state of nature changes. But this simplification can obscure information that may affect our resulting conservation plans. 

Red List Index of species survival for four groups of animals – corals, mammals, birds and amphibians. Source: IUCN

There are several things that are important to think about when choosing how to present indicators: the target audience, the information to show, and the traits of the indicator and data. In a paper recently published in Conservation Biology, my team and I outline the thought-process of deciding what information to show and which method is most appropriate to do so (Figure 1). We used the Red List Index, Living Planet Index and Ocean Health Index as examples to reveal the suitability of the different methods. 

Who is the audience?

Indicators can be used to communicate the state of nature to audiences ranging from the general public to scientists or policy makers. The target audience can affect the most appropriate way the indicator is presented and what information is most important to show. Audiences can have different points of interest and different levels of understanding. For instance, the public may only need to see the trends in an indicator, whereas scientists may be more interested in the technical details associated uncertainty in indicator values.

What information are you trying to convey?

There are two key types of information that we may want to present alongside indicators: 1) the degree of variability in the data used to calculate the indicator or 2) uncertainty in the indicator trend. This information can be presented via intervals, such as confidence intervals. But there are many different methods to do this.

1. Measures of variability can show whether there are patterns in the data – are consistent trends, variation, or strongly uneven trends among observations used to calculate the indicator? Four common methods to measure variability include the standard deviation, median absolute deviation, mean absolute deviation and quantiles, each of which can show different information. 

The more clustered datapoint are in the graph, the less variability there is in the underlying data used to calculate the indicator.

2. Measures of uncertainty can show how reliably the indicator values represent the broader population of interest when the indicator is only calculated using a subset of data. Three commonly used methods for estimating uncertainty are the standard error (used to calculate 95% confidence intervals), bootstrap resampling method and jackknife resampling method.

The wider the intervals, the less certain we are that the indicator value is representative of the current state of nature. Source: Pedro Velica

Indicator and data traits

The traits of the indicator’s formula and data can affect which of the above methods of calculating intervals is most suitable. Some methods are only robust when the data have certain traits (e.g., large sample size, normal distribution). Similarly, some methods are more suitable for indicators with certain types of formulas (e.g., arithmetic mean, geometric mean, sum).

Example using the Living Planet Index

The Living Planet Index (LPI) shows the rate of change in vertebrate species population abundances globally. We calculated the index across terrestrial birds, mammals, and amphibians in the Neotropics. The spread of annual rate of change in population size for each species was often highly skewed, so the methods that assume the data are normally distributed (like a bell curve) were not suitable for showing variability (including the standard deviation and mean absolute deviation) or uncertainty (including the standard error). When intervals were calculated using these methods, the intervals misrepresented the data, such as the intervals impossibly declining to below zero abundance.

The methods that were suitable for showing the level of variability in the annual rate of change among species in each year were the quantiles and medical absolute deviation (Figure 1). These two methods show different information as the quantiles show the middle 50% of the data, whereas the median absolute deviation shows the median of the absolute differences between each observation and the index value for a given year. Jackknifing or bootstrapping were most appropriate for representing uncertainty, and produced similar results (Figure 2).

Figure 1. Four common methods for showing variability in the data used to calculate the Living Planet Index. The wider the intervals, the more variation there is in the data.

Figure 2. Three common methods for showing uncertainty in the Living Planet Index. The wider the intervals, the less certainty we have that the indicator value is representative of that population of interest.

Important findings

The reliability of biodiversity indicators is affected how they are presented. The method used to show variability or uncertainty can vastly alter the key messages conveyed by the indicator. Careful methods selection is essential to ensure biodiversity indicators are conveying the information we are interested in and show it in a reliable way.