Exploring the relationship between

running and heart rate

With the learning of mathematics, I have always been absorbed by its integrations into

everyday life. Being someone looking to go into the field of medicine, it really strikes my

curiosity to find out how I could use math in order to predict likely outcomes through

graphs and correlations. It grabbed my attention when I recently looked into how I could

use statistics in order to create a model to predict likely outcomes of a bodily variable.

When looking at my own life, one such hobby that would exemplify a predicted bodily

outcome would be running.

As one of my core hobbies, running has been a crucial part in my exercise and

enjoyment of life. Specifically, long distance running without stopping has always

fascinated me through its difficulty, always requiring me to pace myself to avoid the

pitfalls of tiredness and fatigue that comes with long distance running. Often, heart rate

can be linked with the level of strain that the body experiences, and is an important

dependent variable for runners to monitor when running at faster paces (the speed at

which one runs at) for longer periods of time. With this in mind, heart rate’s relationship

to pace is exemplified by the need to maximize challenge while keeping risk to a

minimum. As a result, my rationale is that being able to utilize a simple mathematical

equation to figure out the heart rate achieved at a certain pace is crucial to determine

the level of challenge that is suitable for me. My goal for this investigation is to

mathematically determine if there is a relationship between the pace at which one runs

to heart rate. Ultimately, the final goal would be to find a mathematical equation to

predict the heart rate increase in relation to the pace at which I run at.

1
Aim and approach

As indicated in the introduction, the aim for this paper will be to mathematically

determine an equation for which my heart rate is affected by the pace at which I run.

In order to achieve this, I will need to collect data from an experiment, as there is no

data on the internet that is able to create an accurate data set of my heart rate in

homeostasis while running at different paces. For that reason, I will be needing to

conduct live experiments in order to collect data to reach a conclusion.

To specify, heart rate will be measured during the homeostasis of running at a specific

pace, which means that the heart rate has adjusted to supply the extra oxygen to the

body’s increased demand while running at a certain pace. This thus indicates that there

should be a time buffer present before the heart rate measurements are taken.

As a means to approach the aim, I will highlight the independent variable as being the

pace at which I run at and the dependent variable to be my heart rate during

homeostasis in order to determine the relationship as an equation. Thus I need to

create a plot for pace vs heart rate.

I will be utilizing the treadmills at my YMCA in order to standardize, control, and

measure the paces for which I will be running at so that I can collect data. As well, I will

be utilizing a xiaomi band 5 in order to record the heart rate data down.

2
My first thought was to run for around 5 minutes and record a singular point of data for

the pace. However, I realize that there are many variables that sway the recording of

heart rate, and could cause an outlier to occur. In order to fix that, I would be recording

a data point for every 15 seconds, and can decide to average out the recordings in

order to reduce the effect of outliers upon my data. Furthermore, I realize that

increasing pace is continuous, and I would have to have a trial for each speed the

treadmill could precisely change, creating an almost impossible and arduous amount of

time and effort in order to satisfy the data collection. As a result, I would change from

continuous data to discrete data by doing a select range of paces by going up a certain

amount until I reach a maximum as a means to lower the amount of trials I would have

to do in the end.

From previous experiences and knowledge, I would anticipate that the heart rate would

increase to match the level of strain the body had to endure, thus increasing whenever

the pace of the running increased. At faster paces, I would have to slow down faster,

suggesting to me that there will be a non-linear positive relationship between pace and

speed, as the strain upon the body is seemingly exponentially amplified until one cannot

run any faster. This further suggests that the equation’s slope of heart rate increase

increases the faster one runs. In addition, the y-intercept of the equation is the resting

heart rate, allowing for the easy collection of this important data point.

3
My hypothetical expected equation would be an exponential equation likened to

𝑐𝑥
𝑦 = ℎ(𝑏) + 𝑎, where 𝑦 is the average heart rate in the last minute after four minutes of

running, 𝑥 is the pace at which I run at with ℎ 𝑏 and 𝑐 being the factors I need to

manipulate in order to stretch and dilate the graph and 𝑎 being used to translate the

graph up in order to fulfill the asymptote requirement for a resting heart beat.

As a means to test my hypothetical equation, I ran the experiment. The data itself would

range from 7 kilometers an hour (km/h) to 14km/h as that is the speed range for which I

usually run at with the minimum of 7km/h being used to cooperate with the time

constraint of the experiment while the 14km/h would be the fastest I could run on the

treadmill safely. As well, I had decided to go up by 0.5 kilometers with each trial,

generating a good amount of 15 trials and thus data points because it was a good

medium between data fidelity and time investment. In addition, an extra point at x=0 will

be recorded, as that would constitute the heart rate at rest and being a y-intercept to

help determine the type of graph.

As well to clarify, I will be rounding to two decimal places, as any more would be

redundant as bpm could only be measured to a single digit.

Experiment set up, data collection and results

For my data collection, the variables that I will need to collect data for is heart rate and

pace. For the dependent variable, heart rate, I will utilize my friend’s mi-smart-band-5

smartwatch and its function to monitor heart rate in order to record heart rate data. As

4
well, I will have a friend record the heart rate every 15 seconds by telling them what it is

at. All of this will be done running on a treadmill for 5 minutes or 300 seconds in order to

standardize the amount of time and paces that I will be running at. Furthermore, 5

minutes is around the average amount of time that it usually takes to reach the

increased heart rate, with the slope for heart rate’s increase usually dropping off to zero

by around four minutes. For the resting data point, I will take three days of rest once all

the data points have been completed and record the data for a minute, with each point

having a 15 second interval between them to match my experiment. Then the four data

points will be averaged out to get the y-intercept. Some controlled variables that I must

account for are the watch I am using, the placement of the watch, the model of the tread

mill, the incline of the tread mill being flat, and the intervals for which I am recording

heart rate. As well, some inconsistencies in the data can be attributed to the

confounding variable that is my body, as it can be fatigued after certain tests or be

under strain that can affect the heart rate measurement. One way to counteract this

confounding principle would be to take 30 min breaks in between each trial. As well, I

found that the watch’s heart rate monitor can be quite fiddly, and may be a confounding

factor that can affect the accuracy of the test. As a result of these confounding factors,

the data may have outliers, and can mislead the final resulting conclusion.

Unfortunately, due to time constraints, only a single trial can be run for every pace. As

well, due to the large data table, the raw data will be included in the appendix.

After doing the data collection, I had to redo the slowest three, as they had abnormally

high numbers compared to the rest. As well, due to the time constraints, I had to do 9

5
trials in a day, resulting in some outliers that had higher heart rates in the paces

11.5km/h, 12.0km/h, 12.5km/h, 13.0km/h.

In the following table, I have calculated the average heart rates for the last minute of the

experiment using the formula (𝑎1 + 𝑎2 + 𝑎3 + . . . 𝑎𝑛)/𝑛 with 𝑎 being the heart rates

added together and 𝑛 being the number of heart rates that are being added up. This will

reduce the ability for an outlier to affect the data, and will result in more consistency

throughout the graph

The first calculation of 7km/h is demonstrated by inserting the last four recordings of

heart rate corresponding to the last minute of running.

= [(131) + (129) + (132) + (131)]/(4)

= 523/4

= 130. 75

Table for average heart rate in bpm in the last minute vs pace

Table 1: The average heart rates (BPM) for the last minute at each pace

6
From here, I could create a scatter plot graph and analyse the trend

Fig 1: Scatterplot graph of the Average heart rate for the last minute vs pace (excludes resting data point)

As well, I was able to calculate the resting heart rate to be around 67.5

Fig 2: table for the graph averaging of the resting heart rate

Modelling and mathematical interpretation of the data set

In order to determine the best form of graph for the data, I would need to utilize my

mathematical prowess in order to create different equations based on different models

and compare the r-squared values of each form to determine the best model. For my

case, I want to explore the differences between exponential and linear functions to

model my data, and ultimately choose one to determine what the relationship of the

7
data is. I will limit the domain of the equations to be [0, 14] in following with the context

of the situation as we would not be able to achieve a negative speed as well as the

limiting fact that I am only able to run 14km/h on the treadmill.

One way in which to determine the quality of an equation to data is to approximate the

line of best fit for that type of graph and determine its r-squared value for non-linear

data. The r-squared value is a number that quantifies the variation caused by the

relationship of two variables for a specific equation. In this case, the value would be

explaining the percentage variation that running at a faster pace causes heart rate to

increase or decrease. Thus this would give a quantifiable quality to which model best

describes the relationship. The r-squared calculations will be done in the raw data and

appendix portion of the paper to avoid unnecessary calculations.

Investigating and personal creation of lines of best fit

Exponential model

To start, I will first investigate an exponential curve, as that was my initial hypothesis for

an equation when speculating on the results. To start, we will be using the previously

𝑐𝑥
stated equation of 𝑦 = ℎ(𝑏) + 𝑎 in order to represent our data. In context of the

situation, 𝑦 would represent the average bpm in the last minute, 𝑥 being the pace at

which I run at, ℎ, 𝑏 and 𝑐 being variables we can manipulate to fit our graph and 𝑎 being

to adjust our graph’s asymptote via vertical translation. However, for the sake of

simplicity, we will be setting the variable of 𝑏 to two, as well as the value of ℎ to one in

8
order to both reduce the variables I would need to calculate for to just being 𝑐 and 𝑎,

and to have the y-intercept be easily be modelled as 𝑦 = 𝑎 + 1.

To start, I first determined whether or not this was a positive or negative correlation. By

referring to fig 1, I determined that there would be a positive slope and thus a positive

leading coefficient. In addition, we can get the 𝑎 value of the equation as being 66.5 by

subbing the value of the y-intercept into the y of the equation.

67. 5 = 𝑎 + 1

𝑎 = 66. 5

𝑥
Then, using the evidence previously gathered I plotted the equation of 𝑦 = 2 + 66. 5

due to it being able to fulfill the positive slope requirement and going through the x-axis.

The equation produced the following graph.

Fig 3: First attempt at creating an exponential equation

9
When looking at the graph, it is apparent that the model created did not fit the graph due

to the slope’s magnitude being too large. Thus I would need to manipulate the slope by

determining a proper 𝑐 value to the equation. To do this, I would need to choose a point

for the graph to pass through, sub the x and y-values of the point into the equation,

isolate 𝑐 and sub the c value back into the equation to achieve a more accurate

representation of the data. The point I chose was (12,175), as that seemed like a

reasonable point to have the slope of the graph pass through. The following equations

are to solve for the variable 𝑐.

𝑐(12)
(175) = 2 + 66. 5

𝑐(12)
108. 5 = 2

𝑙𝑜𝑔2108. 5 = 𝑐(12)

(1/12)𝑙𝑜𝑔2108. 5 = 𝑐

(1/12)(67. 761551232444) = 𝑐

𝑐 ≈ 0. 56

(0.56)𝑥
Thus subbing the value of 𝑐 into our previous equation we get 𝑦 = 2 + 66. 5 and

producing the following graph.

10
 Fig 4: second attempt at creating an exponential representation of the data

Inferring from this graph, although the graph passes through both the y-intercept and

the picked point, it under estimates the values left of the point and over estimates the

values to the right of the point. In order to fix that, I would have to manipulate the

leading coefficient by making it smaller. However, that would change the c value as well,

and would include two variables I would have to solve for, making it so that it would be

impossible to just isolate a single variable to solve for. In the end, we ended up with a

relatively terrible r-squared value at 0.16, thus denoting that the graph only explains

16% of the variance between pace and heart rate.

Another method we could try would be to estimate a curve by following the parameters

and evidence that the data suggests, and try to tweak different variables to that match

11
the data. Using the evidence we gathered from figure 3 and 4, we can again start off

𝑥
with 𝑦 = 2 . First, I realized that the growth of the slope was too fast, and realized that I

needed to adjust the 𝑐 value to be smaller. Furthermore, while its growth needed to be

culled, I realized that to balance out that change I would need some other factor to

counteract and raise the graph’s slope without affecting the growth of the slope. Thus I

deduced that I needed an ℎ value that was larger than one. From there, I adjusted the ℎ

and 𝑐 values through said trial and error and stumbled across a curve that looked like it

(0.1)𝑥
could fit the graph. At that point, I had the equation 𝑦 = 85(2 ) producing a graph

like this.

Figure 5: The graph of creating an exponential curve through trial and that matches my data.

I then deduced that I would have to shift the graph down, and I decided to translate it

down to go through the y-intercept as that would guarantee it passes through at least

12
one point. To get the vertical shift downwards, I will subtract the value of the y-intercept

of the graph from the value of the y-intercept from the data and sub that into the 𝑎 value

to get the vertical shift downwards.

𝑎 = (67. 5 − 85)

𝑎 =− 17. 5

(0.1)𝑥
All together, we get an equation of 𝑦 = 85(2 ) − 17. 5 that produced the following

graph.

Figure 6: Graph of the final equation with the use of trial and error to produce an exponentially modelled graph

As well, it has an r-squared value of 0.65, a 40% increase in explaining the relationship

between heart rate and pace compared to the previous attempt at creating an

exponential equation.

13
Linear model

Another model that looks like it could fit the graph would be a linear equation. To model

this, I will be using the equation 𝑦 = 𝑚𝑥 + 𝑏, where 𝑦 would represent the average

heart rate in the last minute, 𝑥 being the pace at which one runs at and 𝑏 being the

vertical translation that the graph will have. To start the creation of the graph, we can

immediately acquire the 𝑏 variable as being 67.5, as the value would equate to the

y-intercept which is also the average resting heart rate. The other variable we will need

to calculate is the 𝑚 value, which will be the value for the slope and how pace changes

heart rate. The 𝑚 value can be calculated using the formula 𝑚 = (𝑦 − 𝑦1)/(𝑥 − 𝑥1),

where 𝑥 and 𝑦 are the coordinates of a point on which the linear equation will pass

through and 𝑥1 and 𝑦1 being a different point. For one point, we are utilizing the

y-intercept of (0,67.5), as we want the equation to pass through the y-intercept. For the

other point, by referring to Fig 3, we can see that the point that would create a line best

suited to the graph would be (12,175), as that would result in a pretty even split between

the points on both sides of the equation. The following calculations is to calculate the 𝑚

value of the equation.

𝑚 = (67. 5 − 175)/(0 − 12)

𝑚 = (− 107. 5)/(− 12)

𝑚 ≈ 8. 96

14
Subbing the 𝑚 and 𝑏 values into the model, the final equation for a linear model would

be 𝑦 = 8. 96𝑥 + 67. 5, giving the following graph.

Fig 7: Graph of the linear model for the relationship between pace and heart rate

I personally like this graph, as it is pretty in the middle of all the data points as well as

looking like it passes through five of my data points. As well, when calculating the

r-squared value of 0.86 or explaining 86% of the variance between the pace and heart

rate, it was the largest out of all the r-values so far discussed within this paper. Overall,

this surprised me as being a better fit than my initial hypothesis of it being exponential.

Technological verification of models

To end off my investigation, I will be verifying the two models that I have created using

technology. Specifically, comparing the r-squared values of my models to the computer

15
generated models as well as key points such as the y-intercept and ultimately decide on

the equation that best represents my data.

Exponential model

I plugged in all the values from my table including the resting value into the calculator

and utilized the exponential regression function to find the line of best fit for an

𝑥
exponential model. The equation for the graph was 𝑦 = (77. 22646)1. 06931 and had

the highest r-squared value so far of 0.92. Interestingly, there was an absence of a

translation present, unlike the route I took while creating the first exponential model. As

well, it has a y-intercept of 77.2, indicating that the resting heart rate would be 77.2

bpm, which is off by 9.7. Furthermore, its small base value suggests that the data is

more suited to a linear equation rather than an exponential one like I had first

hypothesized. Overall, I personally think that the computer generated equation is the

best representation of an exponential model, as it has the highest r-squared value even

if it overestimates the resting heart rate.

Linear model

For verifying my linear equation, I utilized the linear regression function to find the best

fit for the data. Utilizing the linear regression on my calculator gave me an r-squared

value of 0.94, a slightly higher r-squared value than the best fit for an exponential

model. As well, its y-intercept was at 72.7, closer to the resting value of 67.5 and

beating the exponential model by five units. As well, due to its nature as a linear

equation, the r value is able to be used as the correlation coefficient, describing both the

strength of the equation through its magnitude being closer to 1 and whether it's a

16
positive or negative relationship. With that in mind, judging from the r-value of 0.972, we

can see that its magnitude is quite close to 1, revealing that there is a strong correlation

between the heart rate and pace. As well, the r-value being positive rather than negative

suggests that the data is positively correlated, and thus we can conclude that the

relationship between heart rate and pace is strongly positively linearly correlated. As for

comparing it to my own linear regression, its high correlation suggests that it is indeed a

more accurate predictor for my heart rate. However, when wanting the resting heart

rate, my linear line of best fit would be more accurate, and thus diminishes some of its

significance.

Evaluation, reflection, and conclusion

Evaluating the best equation, we can first eliminate the possibility of it being an

exponential equation, as its r-squared value for the computer generated best fit is lower

that the computer generated best fit thus proving that the data follows a linear

relationship. As such, that leaves us with two possibilities, my own personally created

graph and the computer generated graph. Between the two, I chose my equation

𝑦 = 8. 96𝑥 + 67. 5 as the better predictor for heart rate, as although it’s r-squared value

was a bit smaller, it had a more accurate predictor for the resting heart rate, a value that

I would deem to be the most important due to the accuracy in its acquisition.

To reiterate, the purpose of this paper was to evaluate the relationship between heart

rate and pace at which I run at, and to ultimately determine a mathematical equation to

estimate heart rate’s value at different running paces. With that being said, I can

17
conclude that the data best follows a positive linear correlation, and has resulted in me

being able to create an accurate equation that I would deem as an accurate

representation of my heart rate. Although it did not match my hypothesis of the

relationship being exponential, I was still able to reach an equation that could best

model the relationship, thus answering the aim that I had set out to achieve.

However, this paper did not come without regrets and limitations. For one, I am missing

a large amount of data relating to the middle between 0.5km/h and 6.5km/h, as the time

constraint for this experiment restricted me from gathering more data. As well, the

confounding variable that is my body played a major role in the experiment’s results, as

I was sick on the break that I had decided to gather the data, resulting in there being a

month gap between when some of the data was collected. Another issue was that the

time constraint and situation had me gathering 9 data points in a day, thus resulting in

the abnormally high number seen in the recordings from 10km/h to 13km/h. Lastly, I

would have liked to include a quadratic equation to see if it matched my data, but I was

limited by the amount of pages that I could have on this paper.

In conclusion, the pace at which one runs at follows a direct linear relationship with

heart rate, and can be modelled best using the equation 𝑦 = 8. 96𝑥 + 67. 5.

18
bibliography:

-​ (Graphs and computer generated regressions were done on) Desmos | Beautiful

free math. (n.d.). https://www.desmos.com/

-​ (regression formulas used)

Numeracy, Maths and Statistics - Academic Skills kit. (n.d.).

https://www.ncl.ac.uk/webtemplate/ask-assets/external/maths-resources/statistics/regressi

on-and-correlation/simple-linear-regression.html

-​ (Heart rate increases as you run faster, and good estimate for workload on body)

Chertoff, J. (2024, February 8). What’s my ideal running heart rate? Healthline.

https://www.healthline.com/health/running-heart-rate#:~:text=During%20aerobic%20exe

rcise%20like%20running,how%20hard%20you%27re%20working

19
Appendix:

Raw data and r value calculations

Raw heart rate data for each pace:

20
Regression Calculations for Each Graph

21
22