Many different coloured flowers, but which are attracting the pollinators?
Home ยป Scientists Put 5 Different Ways To Calculate Pollinators’ Favourite Flowers To The Test. The Best Is Simple Enough To Apply To Your Own Garden.

Scientists Put 5 Different Ways To Calculate Pollinators’ Favourite Flowers To The Test. The Best Is Simple Enough To Apply To Your Own Garden.

Scientists found that methods for quantifying bees’ floral preferences fundamentally disagree on which flowers bees find most attractive. In a new article, they identify the most reliable method, one that’s simple enough to apply to your own garden.

Plants are incredibly important for supporting pollinators such as bees, butterflies, and other beneficial insects. When individuals or organizations aim to help pollinators by planting butterfly gardens or wildflower areas, it can be difficult to know which flower species deliver the most bang per buck. Rachel Pizante and colleagues from the University of Alberta evaluated various methods scientists have developed to quantify pollinators’ flower preferences and provide planting recommendations. However, they discovered major inconsistencies among these techniques, which brings their reliability into question. Their study, published in the journal Insect Conservation and Diversity, found that different mathematical techniques for calculating pollinators’ flower preferences do not agree on which flowers are most attractive.

The challenge of determining pollinator preferences

Researchers have tried to determine which flowers pollinators prefer by looking at which flowers pollinators visit most often. However, the number of visits alone can be misleading. Pollinators might just be visiting the most abundant flowers in an area, not necessarily their favourite flowers. The inherent abundance of different flower species needs to be accounted for.

So, researchers have developed mathematical methods to calculate pollinators’ true flower preferences while removing the effect of flower abundance. These methods are called “preference metrics”. They work by comparing the number of visits to a flower species compared to how abundant that species is. The metrics calculate a preference score for each flower that shows if pollinators visited it more than expected based on its abundance. This allows researchers to rank which flowers pollinators seem to prefer most.

Shocking disagreement between methods

Shockingly, when the researchers tested these different preference metrics on the same dataset, the metrics gave very different rankings for which flowers were most preferred! The metrics fundamentally disagreed on which flowers pollinators found most attractive.

A heat map in shades of blue. If there were a consensus between models, then the map would appear as a series of stripes. In fact it's just a mess of blocks.
Figure 1 in the paper: Heatmap showing the ranks given to each flower species by each metric. The most preferred species is given the rank of 1, while the least preferred species is given the rank of 35. CI, confidence interval; MAH, mass action hypothesis; PI, preference index. Source Pizante et al. 2023. Click to enlarge.

One example is Symphoricarpos occidentalis, Wolfberry. Which Pizante and colleagues single out as an example. They write:

There was no consensus among metrics regarding which flowers were most preferred (Figure 1). In fact, different metrics listed the same flower species as preferred (highly ranked) and avoided (lowly ranked). For example, Symphoricarpos occidentalis was considered most preferred by the mass action hypothesis and confidence interval metrics, somewhat preferred by the centrality metric, and not preferred by the resource preference and PI metrics (Figure 1). Additionally, even when some species showed similarities between metrics, these same metrics could rank other species quite differently. For example, although the mass action hypothesis and confidence interval metrics produced the same rank for S. occidentalisSolidago canadensis was ranked third by the mass action hypothesis metric and 34th by the confidence interval metric.

Pizante et al. 2023

This disagreement stems from the fact that the metrics differ in how well they account for the abundance of each flower species. Some methods remove the effect of flower abundance better than others. The better a metric controls for abundance, the more it reflects pollinators’ true innate preference rather than just availability.

The metrics also relate differently to simply using the number of pollinator visits to rank preference. The study found that some metrics are correlated with visit counts, while others provide novel information. For example, the researchers found that the “preference index” metric showed no correlation with visit number. This suggests it provides additional insight into preferences not contained in visit data alone. In contrast, the “mass action hypothesis” metric was strongly positively correlated with visit counts, indicating it does not give much new information beyond raw visit data.

Tracing pollinator steps through Alberta’s prairie wildflowers

In this research, the scientists tested five different preference metrics on the same dataset of flower visits collected in the prairie regions of Alberta, Canada. The prairies are an iconic landscape in western Canada, covered in grasses, wildflowers, and shrubs. These habitats support diverse pollinator communities.

The researchers observed and recorded flower visitors along transects in mixed-grass prairie at the Mattheis Research Ranch. They counted and identified 268 pollinator species that visited 35 different flower species native to the prairies. The most common pollinators were bees, flies, butterflies and beetles. The team also counted the number of flowers of each plant species.

The study provided a robust dataset to evaluate the preference metrics, encompassing the interactions between prairie pollinators and naturally occurring flower species. The metrics could then be calculated and compared using the same underlying visitation and abundance data. This allowed the researchers to reveal the inconsistent results among techniques intended to determine pollinators’ floral preferences.

Demystifying the preference calculations

The preference metrics each calculate a mathematical score for how preferred each flower species is, using the visit data and abundance data. But they go about it in different ways:

  • The “confidence interval” and “resource use” metrics simply look at whether a flower got more visits than expected based on its abundance. For example, a common flower would need lots of visits to be considered preferred by these metrics.
  • The “preference index” and “mass action hypothesis” metrics use the relative proportion of visits compared to abundance. This retains more information about the flowers’ abundances.
  • The “centrality metric” ignores abundance entirely. It only looks at the structure of connections in the interaction network between pollinators and plants.
  • Each metric uses its score to rank all the flower species from most preferred to least preferred.

The problem is, when tested on the same data, these different approaches generate inconsistent rankings of the flowers. The researchers found that no two metrics strongly agreed on which flowers were most attractive to pollinators.

Selecting the optimal metric

Based on their analysis, the researchers recommend using one metric called the “preference index” over the others. They highlight two key reasons:

First, the Preference Index was best at removing the effects of flower abundance to uncover true pollinator preferences, say Pizante and colleagues. They write:

The PI metric assesses the relative number of visits a plant species receives versus the relative number of flowers of that species, therefore retaining information regarding relative abundance. We found that the PI metric is significantly negatively correlated with flower abundance but showed no correlation with the number of visits.

Pizante et al. 2023

That lack of correlation means that the Preference Index provides more information than just the number of pollinator visits. They write:

When flowers are rare, the denominator of the PI equation will be very small, such that few visits will result in the numerator being large enough for a plant to be considered preferred. The opposite is true as well such that a common flower will result in a large denominator and will require a large numerator to be considered preferred. We expect an insignificant correlation with the number of visits because rare plants can still be highly preferred even though they usually have few visits relative to the total number of visits to all flower species in the dataset. An example of this in our dataset is Escobaria vivipara, which only had five flowers, but received 11 visits. We would therefore expect this plant to score a high value from a given preference metric, which the PI metric accomplishes, and for the metric to not be correlated with the number of visits. Thus, the PI metric functions in the way we would expect a working preference metric to function.

Pizante et al. 2023

Another advantage they give for the Preference Index is that it is insensitive to undersampling, which means you don’t need large numbers of plants for it to produce useful results. Most importantly for the home gardener, it’s relatively simple to use. If you have the patience to spend a morning watching insects buzz around your flowers, you might get some interesting results. The full details are in the Open Access paper, which would allow you to discriminate by pollinator species. If you’re happy with any pollinators visiting your garden, then a slightly simplified method is below.

Using the Preference Index in your garden

If you want to use the Preference Index in your own garden, here is an overview of how it works:

  1. Identify the flower species in your garden and count how many flowers of each species you have. This gives you the abundance data.
  2. Observe pollinator visits to the flowers over time, recording each time a pollinator visits a particular flower species. This gives you the visitation data.
  3. For each flower species, divide the number of visits to that species by the total visits to all flowers. This gives you the relative visit proportion for that species.
  4. For each flower species, divide the number of flowers of that species by the total flowers of all species. This gives you the relative abundance.
  5. Divide the relative visit proportion by the relative abundance for each flower. The resulting number is that flower’s preference index.
  6. Rank all the flowers by their preference index. Those with the highest scores are the most preferred by pollinators in your garden!

Pizante, R., Acorn, J.H., Worthy, S.H. and Frost, C.M. (2023) โ€œExisting flower preference metrics disagree on best plants for pollinators: which metric to choose?,โ€ Insect Conservation and Diversity. Available at:

Alun Salt

Alun (he/him) is the Producer for Botany One. It's his job to keep the server running. He's not a botanist, but started running into them on a regular basis while working on writing modules for an Interdisciplinary Science course and, later, helping teach mathematics to Biologists. His degrees are in archaeology and ancient history.

Add comment

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Read this in your language

The Week in Botany

On Monday mornings we send out a newsletter of the links that have been catching the attention of our readers on Twitter and beyond. You can sign up to receive it below.

@BotanyOne on Mastodon

Loading Mastodon feed...