# Birds Don’t Check The Temperature Before Visiting The Feeder

June 22, 2021For the *Cornell Feeders Live investigation*, we have shared visualizations that highlight (1) the sampling effort (i.e., amount of time watched) and (2) when the study species visited the feeding station. This second set of visualizations helped us start to answer our first research question: What is the daily visitation pattern of different species at the feeders?

Today, we want to share visualizations that help us answer our second research question: How does weather affect the probability of different species visiting the feeder? More specifically, how does temperature affect the probability of different species visiting the feeder? While the community was interested in understanding the effects of precipitation, there was not enough precipitation data (i.e., most hours had zero inches of precipitation).

We created visualizations after running statistics that assessed not only the potential effect of temperature, but also the potential effect of time of day and sampling date. Time of day and sampling date could both be influential when trying to understand behavior at the feeding station.

Each visualization explores how each predictor (temperature, time of day, sampling date) relates to the percentage chance that a species is present at the feeding station. To explore each of these relationships, we allowed one predictor to vary and held the other predictors at their average value. For example, to explore the effect of temperature across its range, we held time of day and sampling date at their average value. If you are interested in learning more about how we did this, jump down to the section “Behind The Curtain.” Otherwise, read on to check out the visualizations.

**What did we find?**

Even though there was substantial variation in temperature, ranging from 18 to 75 degrees Fahrenheit, temperature was not related to any of the eight species being at the feeding station. The presence of all eight species was relatively the same regardless of temperature (Figure 1). Even though the percentage chance of some species, like the American Goldfinch, seems to change with temperature, the 95% confidence bands show that there really isn’t a statistically different change. The bands indicate where the line itself could vary and still be considered the same.

**Temperature does not affect the percentage chance a species is present **

*Figure 1. The estimated percentage chance a species will arrive at the feeding station by temperature with 95% confidence bands (i.e., range in which we are 95% sure lies the true value). Each point represents the percentage chance of a species arriving during that temperature, holding all other predictor variables at their average. Each species has its own color, and its data can be toggled on and off.*

Species presence at the feeding station depended more on the time of day and sampling date. For seven of the eight species, the chance that a species would be present at the feeding station increased as the morning advanced and then decreased as the day drew to an end (Figure 2). However, the shape of this rise and fall of the percentage chance seems to depend on the species. For example, for the Red-winged Blackbird the percentage chance was very high for the majority of the day until the last couple hours, while the Blue Jay was relatively high earlier in the first half of the day and then dramatically fell in the second half of the day.

**The percentage chance of a species being present depends on the time of day**

*Figure 2. The estimated percentage chance a species will arrive at the feeding station by hour (5:00 A.M. to 7:00 P.M.) with 95% confidence bands (i.e., range in which we are 95% sure lies the true value). Each point represents the percentage chance of a species arriving during that hour holding date and temperature at their average. Each species has its own color, and its data can be toggled on and off.*

For the American Goldfinch, White-breasted Nuthatch, and Tufted Titmouse the percentage chance of being present at the feeding station also depended on the sampling date (Figure 3). As the sampling period advanced from March 31 to April 14, the percentage chance of these three species decreased.

**The percentage chance of a species being present decreases with sampling date for three species **

*Figure 3. The estimated percentage chance a species will arrive at the feeding station by sampling date (March 31 to April 14) with 95% confidence bands (i.e., range in which we are 95% sure lies the true value). Each point represents the percentage chance of a species arriving, holding the other predictors (temperature and hour) at their average. Each species has its own color, and its data can be toggled on and off.*

**Behind the curtain: How we created the visualizations**

Before we ran the statistical analysis, we first converted the raw observations into a presence-absence data set. For each hour watched from 5 A.M. through 7 P.M., we assigned a “1” if a species was seen and a “0” if a species was not seen at the feeding station. Then, for each of the eight study species, we ran a model (specifically, a binary logistic regression) with temperature as the weather variable as well as predictors “hour” and “sampling date,” two other important variables that we expected to be related to species presence at the feeding station.

To determine the statistical significance of each predictor variable, we used Wald Z-tests and resulting p-values. If the p-value of a predictor variable was less than 0.05, then we considered it statistically significant. In other words, we considered a predictor statistically important when there was less than a 5% chance the relationship between the predictor and the response (presence of a species) was by random chance alone. Only then we were comfortable saying it was meaningful at the population level (learn more about p-values and regression).

To visualize the effect of each predictor variable, we calculated their average value (temperature = 51, Hour = 12, date = April 7) and then used the model to calculate predicted values. For example, to visualize the relationship between temperature and a species being present at the feeding station, we allowed temperature to vary (25 to 75) and then held the other predictor variables at their average.

**What do you think?**

Share your interpretations of the data and your questions in the forum below. What do you think is driving these patterns that we are seeing? Why is temperature seemingly not important? What do you think could explain the differences between species?

We invite everyone’s input in this phase of the investigation (data exploration). We want to hear what you think because your input is key to how we frame our findings in the final report at the end of the investigation.