
Shannon vs Simpson diversity index: which one should you report?

The Shannon vs Simpson diversity index question comes down to which species you want the number to listen to. The Shannon index gives more weight to rare species, while the Simpson index is driven mostly by the common, dominant ones. So the two can rank the same pair of sites in opposite order. For a monitoring report, the safest answer is to show both, next to species richness and evenness, along with how much sampling sits behind them.
This guide is for project developers and community monitoring teams who see these indices on a dashboard and have to explain them to a funder or a verifier. It uses a worked example with two invented sites, so you can check every number yourself.
Shannon vs Simpson diversity index: the short answer
Both indices combine two things: how many species you recorded (richness) and how evenly your records are spread across them (evenness). They differ in how they weigh the two.
Shannon-Wiener (H') | Simpson (1-D) | |
|---|---|---|
What it asks | How uncertain are you about the species of the next individual you record? | How likely is it that two individuals picked at random belong to different species? |
Formula | H' = -sum of p ln p over all species | 1-(sum of p squared) |
Range | 0 upwards, with no fixed maximum (the most it can reach is ln S) | 0 to just under 1 |
Most sensitive to | Rare species, compared with Simpson | Common, dominant species |
Effective number of species | exp(H') | 1 / D |
Here p is each species' share of all the individuals (or records) counted, and S is the number of species.
Morris et al. (2014) compared six common measures on 60 grassland plots from the German Biodiversity Exploratories and could not identify one ideal index. In their notation, D1 is Simpson's diversity (1-D) and D2 is the inverse of Simpson's original index (1 / D). They place the measures on a scale from rare to abundant species: "S is sensitive to rare species, D1 and D2 are sensitive to abundant species", the Berger-Parker index sits at the far end because it responds to only the most abundant species, and H' is "equally sensitive to rare and abundant species". So they suggest H' "could be used in situations where rare and abundant species/traits are expected to be equally important." Their advice is to choose "a priori" a small number of measures and report them together, rather than picking whichever gives the most striking result.

What each index actually measures
Shannon-Wiener index (H')
The Shannon index comes from information theory. For each species, multiply its share of the records by the natural log of that share. H' is the negative of the sum. A community where every record is the same species scores 0. As you add species and spread records more evenly, H' rises.
The log is the reason rare species matter more here than in Simpson. A species with a 5% share still adds a visible amount to H', because ln(0.05) is a large negative number. The vegan documentation notes that natural logs are the most common choice, and some analysts use base 2. Always state the base, because H' values in different bases cannot be compared.
Simpson's index: three versions of one idea
Simpson (1949) described the probability that two individuals drawn at random belong to the same species. That raw figure, D, goes down as diversity goes up, which confuses readers. So two transformed versions are used more often:
1-D (sometimes called Gini-Simpson): the chance that two individuals are different species. Higher means more diverse.
1 / D (inverse Simpson): the number of equally common species that would give the same D.
The vegan package returns 1-D when you ask for "simpson" and 1 / D for "invsimpson". Before you compare your figure with a published one, check which of the three versions the other source used.
Because the shares are squared, a species with a 55% share contributes 0.3025 to D, while a species with 5% contributes only 0.0025. Simpson's index is therefore mostly a statement about the few species that dominate your records.
A worked example: two invented sites
The two sites below are invented for illustration. Each has 100 individuals recorded.
Site A has 10 species: one with 55 individuals and nine with 5 each.
Site B has 4 species with 25 individuals each.
For Site A, H' = -(0.55 x ln 0.55 + 9 x 0.05 x ln 0.05) = 1.68, and D = 0.55 squared + 9 x 0.05 squared = 0.325, so 1-D = 0.675. For Site B, H' = ln 4 = 1.39 and D = 4 x 0.25 squared = 0.25, so 1-D = 0.750.

Shannon says Site A is more diverse (1.68 against 1.39). Simpson says Site B is (0.750 against 0.675). Neither is wrong. Site A holds nine species that Shannon counts in its favour, while Simpson mostly sees one species making up more than half of the records. Site B has fewer species, but none of them dominates.
If a funder only saw one of these numbers, they would draw the opposite conclusion depending on which one you chose. That is the practical reason to report more than one.
Richness and evenness belong next to both
Two simpler numbers make the indices readable.
Species richness (S) is the count of species recorded: 10 at Site A and 4 at Site B. It ignores abundance completely.
Pielou's evenness (J) divides H' by ln S, giving a score from 0 to 1. Site B scores 1.00 because every species has the same count. Site A scores 0.73 because one species dominates. The biodiversity monitoring page explains why this pairing helps restoration projects in particular: early-stage sites "often show improving richness before evenness catches up."
Morris et al. add a caution: evenness derived from H' "mathematically correlates with H'", so J is a way of reading H', not a fully independent measure.
Sample effort moves every number
Diversity indices describe the records you have, not the whole landscape. More effort finds more species, especially rare ones.
Take Site A again. With all 100 individuals counted, you have 10 species. With the standard rarefaction formula (the one the vegan package runs as `rarefy`), a random subsample of only 10 individuals from Site A would contain about 4.7 species on average, and a subsample of 10 from Site B about 3.8. At low effort, a 10-against-4 difference in richness shrinks to less than one species. If one community submits 100 records and another submits 10, comparing their raw indices can tell you more about effort than about diversity.
The CitizenClimate biodiversity monitoring page makes the same point about its own dashboard. It says Shannon and Simpson indices "will underestimate true diversity when sampling effort is insufficient", advises treating indices from fewer than 30 to 50 observations "as provisional", and notes that "The dashboard displays all indices regardless of sample size." It also states that Pielou's evenness "requires at least two species and becomes more stable above S = 5."
Hill numbers: one scale for all three
The neatest fix for comparing indices is to convert them into effective numbers of species, often called Hill numbers after Hill (1973). Richness is the Hill number of order 0, exp(H') is order 1 and 1 / D is order 2. Jost (2006) argues that raw Shannon and Simpson values are not diversities in themselves, and that converting them to effective numbers gives them an intuitive meaning.
The effect is easy to see. A perfectly even community of 4 species has H' = 1.39; one of 8 species has H' = 2.08. Doubling the species adds 0.69 to H', but exp(H') goes from 4 to 8, which is what most readers expect. In the worked example, Site A has 5.35 effective species at order 1 but only 3.08 at order 2. The gap between the two tells you how strongly one species dominates.
Chao et al. (2014) extend rarefaction and extrapolation to Hill numbers, so sites with different effort can be compared at the same sample size or the same sample coverage. Their methods are in the iNEXT package for R.
What to report in a monitoring report
A verifier or funder should be able to recompute your numbers and see how much effort sits behind them. A short checklist:
Report | Why |
|---|---|
Species richness and number of records | Shows the effort behind every index |
Shannon H', with the log base | Weights rare species; base changes the value |
Simpson, stating 1-D or 1 / D | Weights dominant species; three versions are in use |
Pielou's J | Separates evenness from richness |
Survey design: plots, visits, dates, taxa | Indices from different designs are not comparable |
Rarefied or coverage-based comparison when effort differs | Removes the effort effect before comparing sites or years |
The raw species records | Lets anyone re-run the calculation |
Two points apply to community data in particular. First, a record is not always an individual: the same bird can be recorded twice, and a quadrat photo may show one plant or a patch. Decide in your field protocol what one record means. Second, keep effort steady: the same plots, the same visit length and the same season each year. Our guide on how to record bird calls for AI identification shows what a fixed field routine looks like for one taxon.

How the CitizenClimate biodiversity dashboard handles them
The Biodiversity screen described in the App Guide shows totals for species, observations and categories, plus "Diversity indices such as Shannon-Wiener, Simpson's and Pielou's evenness, each with a short explanation". The Data Integrity and Methodology Protocol lists the full set computed "from the raw observation data", including Species Richness, Alpha Diversity ("mean species richness per submission point"), Beta Diversity (Sorensen index) and Gamma Diversity ("total unique species across the entire project"). It also records "Observation counts per species (multiple sightings tracked individually)", so the shares behind each index are shares of observations.
For Simpson, the biodiversity monitoring page gives the dashboard formula as one minus the sum of n(n-1) divided by N(N-1), where n is each species' count and N the total, the version for drawing without replacement. On the worked example that gives 0.682 for Site A and 0.758 for Site B, slightly higher than the 1-D values above.
A few limits are worth knowing:
The platform does not set your sampling design. In the page's words, "The platform does not enforce a specific field protocol", so effort control is the project's job.
The dashboard shows cumulative figures. "Time-series dashboards are on the development roadmap." Until then, the species data can be exported as CSV or JSON and filtered by date, and the Protocol states that exports "include raw records, not just summaries", so you can re-run everything in vegan or iNEXT.
Verifiers can be given a read-only VVB role with direct access to the project data.
The 60-second walkthrough below shows a fictional project, Riverside Forest Restoration, from its first plant records to its diversity indices. The project and its figures are illustrative. The biodiversity survey app walkthrough page goes through the same steps in writing.
For projects under Plan Vivo's PV Nature standard, the Protocol describes Pillar 2, Species Diversity, as "how many species there are and how evenly individuals are spread among them (Shannon diversity)", and Pillar 1 as species richness "adjusted for sampling effort". Under PV Nature, an approved Data Analytic Provider calculates the official pillar metrics. CitizenClimate is not one of them and has no affiliation with the Plan Vivo Foundation. Its role is to deliver structured records and a working view of the metrics, as the PV Nature biodiversity monitoring page explains.
Frequently asked questions
Is a higher Shannon index always better?
No. A higher H' means more species, more evenly spread, in the records you collected. It can rise simply because a team surveyed more, or because a common native species declined while scattered generalists increased. Read it next to richness and evenness, check how many records sit behind it, and compare sites or years only at similar effort, or after rarefaction.
What is a good Shannon diversity index value?
There is no universal benchmark. H' depends on the log base, the taxa surveyed, the habitat and the sampling effort, and its maximum for S species is ln S. A value means something only against a comparable baseline: the same plots, methods and season in an earlier year, or a reference site surveyed the same way.
Should I report Simpson's D, 1-D or 1 / D?
Report 1-D or 1 / D, and name the one you use. Raw D falls as diversity rises, which confuses readers. 1-D reads as a probability between 0 and 1. 1 / D reads as an effective number of species, which sits on the same scale as richness and exp(H'), so it is easier to compare across sites.
How many observations do I need before the indices mean anything?
The biodiversity monitoring page treats indices from fewer than 30 to 50 observations as provisional. That is a rough floor, not a target. More useful than a fixed number is a rarefaction curve: if the species count is still climbing steeply as records are added, the survey has not yet captured the community, and indices will move as effort grows.
Get the numbers and the records behind them
Shannon and Simpson answer slightly different questions, so the most honest report shows both, puts richness and evenness alongside, and says how much effort produced them. A dashboard gives you the figures; the raw records let anyone check them. Download the free app to set up a biodiversity survey, or read the App Guide to see how the Biodiversity screen works.



Comments