Age and Insurance Premiums

How does age correlate with insurance pricing?

Mathematics

Word Count: 2668

 1

Table of Contents

Introduction…………………………………………………………………………………...… 2

Significance and Linear Regression………………………………………………………..... 3

Methodology…………………………………………………………………………………..... 8

Young Age Group…………………………………………………………………………….... 9

Old Age Group………………………………………………………………………………... 14

Conclusion…………………………………………………………………………………….. 18

Word Cited…………………………………………………………………………………….. 19
 2

Introduction

Insurance is an important service to protect people from potential issues,

disasters, or mistakes that occur later in life. There are many different types of

insurances, varying from life insurance (a payout to your family in case of a death) to

fire insurance (payout in the instance of a destructive fire). Usually, a person will pay a

monthly fee, known as a premium, to retain coverage from an insurance provider, given

that the provider will help pay for future repairs or damage as needed. The main type I

will be focusing on in this essay is automobile insurance. Auto insurance covers the cost

of any potential costs incurred while driving. When deciding on a premium, insurance

companies will often hire actuaries to determine the risk of an individual applying for

insurance. Many different factors can affect the cost of insurance such as location,

annual mileage, credit record, and type of vehicle. The higher the risk of a person to

crash, the higher the cost of their insurance.

The mathematical modeling of how insurance prices change based on certain

factors is extremely important to determining the best price for insurance applicants. If a

person is more likely to crash, it is beneficial to an insurance company to charge a high

risk individual more money per month. If a person has a very good driving record, it is

important for an insurance company to charge a low premium so the customer is likely

to stay with the company.

 3

Significance and Linear Regression

In the process of creating a model for age and insurance pricing, it is important to

determine the significance of the discovered relationship between age and insurance

pricing. To do this, the slope of the least squares regression line (LSRL) will be tested in

a linear regression (LinReg) test. An LSRL can be used to make predictions of a

general population based on samples.

1.1: Statistical Significance

A result can be defined as statistically significant when it is very unlikely to occur

given a null hypothesis, H 0 . When an alternative hypothesis ( H a ) is tested against H 0 ,

and the probability p is less than the given significance level ( a ), then H 0 can be

rejected and H a is true. Statistical significance is very important because it tells if a

result can be generalized to a population. If the probability of a result is less than the

significance level, it can be concluded that the sampling error is not responsible for the

results and they can be generalized to the whole population and therefore more useful

for predictions. If the probability is higher than the significance level, then there was not

enough of a correlation to generalize from the sample.

1.2 Errors

There are two types of errors: type I error and type II error. A type I error occurs

when the null hypothesis is true, but it is rejected. This denotes that a false positive has
 4

occurred, that we assert a new claim that is not true when compared to H 0 . To

minimize type I error, an investigation would use a low significance level to limit the

room for detecting a false positive. The other type of error, type II, occurs when H 0 is

false but is not rejected. This would imply that a false negative has occurred, or that the

results were statistically significant but the H 0 was not rejected. To minimize a type II

error, a higher significance level should be used as it will allow for more statistically

significant results to not be rejected.

Since type I and type II error require the opposite conditions to be minimized, the

significance level needs to be chosen in the context of which error is worse should it

occur. Choosing a higher significance level would imply that the type II error would be

worse should it occur, while a low significance level implies that a type I error would be

worse if it occured.

In the context of a company providing car insurance, a type I error would mean

there is no relationship between age and annual price of insurance, but it is concluded

that there is a relationship. A type II error would imply that there is a relationship

between age and the price of insurance, but it is concluded that there is no relationship.

Since a company is the most likely to be doing an analysis of age and annual price, the

point of view of the company providing insurance will be used. For a theoretical

company, a type I error is worse because it means that the company would be giving

estimates that are largely inaccurate and based on the results that were due to random

chance, hurting their reputation.To minimize this, a lower significance level of .05 will be

used.
 5

1.3 Linear Regression Test

A linear regression (LinReg) test determines the significance of the relationship

between two variables. In a LinReg test, the slope or relationship found ( H a : b0 =/ 0 ) is

tested against no relationship ( H 0 : b0 = 0 ) where b0 is the slope of a linear regression

relationship. If the relationship is significant, this means that the equation found is useful

for making predictions about ages that were not observed.

In order to do a LinReg test, certain conditions must be satisfied. The relationship

between the independent variable x and the dependent variable y must be linear and

random. The distribution of the dependent variable y must be normal or approximately

normal and the standard deviation must be constant.

1.4 Equations

These three equations can be used to look for relationships within data:
︿
y = b1 + b0 x
rS y
b0 = Sx

b 1 = y − b 0x

︿
y is the predicted insurance annual cost

b1 is the y-intercept
 6

b0 is the slope

S y is the standard deviation of annual insurance prices

This can be calculated as:

√
n
∑ (y i −y) 2
i=1
Sy= n−1

n is the sample size

y i is each recorded annual insurance price

S x is the standard deviation of age

This can be calculated as:

√
n
∑ (xi −x) 2
i=1
Sx= n−1

x i is each recorded annual insurance price

r is the correlation coefficient

x is the average age

y is the average annual insurance cost

 7

When finding the test statistic

b−β
t= SE b

b is the slope of the sample

β is the slope of the null hypothesis

S E b is the standard error of the sample slope

When finding the standard error of the sample slope

s
SE b = √n
( S1x )

When creating a confidence interval:

b ± t * ( √sn )

n is the sample size

s is the sample standard deviation

t * is the critical value

All above equations from ​The Practice of Statistics.​ Equation of standard error from

“Mathematics of Simple Regression”

 8

Methodology

I will create multiple LSRLs to determine the relationship between age and

insurance pricing. Since the

relationship between age and

insurance is nonlinear, I will

divide the ages into 2

categories: young (16-25) and

old (35-85).

For both of these age

groups, I will perform a LinReg

test to determine the significance of the slope.I will interpret the meaning of the slopes

for each age group and what use they could have.
 9

Young Age Group (16-25)

In the 16-25 year old age group, there is a strong, negative, linear relationship

between age and average insurance price. The LSRL between age and annual

insurance prices can be found:

√
n
∑ (y i −y) 2
i=1
Sy= n−1

√
10
∑ (y i −3567.620) 2
i=1
Sy= 10−1

√
30,503,371.68
Sy= 9

√
30,503,371.68
Sy= 9

S y = 1840.995

√
n
∑ (xi −x) 2
i=1
Sx= n−1

√
10
∑ (xi −20.5) 2
i=1
Sx= 9
 10

Sx=
√ 82.5
9

S x = 3.028
rS y
b0 = Sx

(−.938)(1840.995)
b0 = (3.028)

b 0 =− 570.295

b 1 = y − b 0x

b 1 = 3567.620 + (570.295)(20.5)

b 1 = 15258.668
︿
Which leads to an LSRL of y = 15268.668 − 570.295x , as shown below with

domain limit 16 < x ≤ 25 .

The correlation coefficient can be calculated as follows:

 11

r = .938

r 2 = .8798

88.0% of variation in insurance pricing is explained by age. For each additional

year of increased age, the predicted average cost of insurance decreases by about

$570.30.

The residual plot shows a

pattern as age increases,

meaning the LSRL is not

necessarily appropriate. This

could be due to external factors

affecting the price of insurance that are not related to age such as gender and type of

vehicle. As there is not many as many factors to consider for younger individuals, age is

acts as an important factor explaining car insurance pricing, reflected in the high r 2

value. Although the line is not necessarily accurate, a test will still be done, with the

possible external factors kept in mind.

The null hypothesis is there is no relationship between age and insurance pricing

( H 0 = 0 ). The alternate hypothesis is that there is a relationship between age and

insurance pricing (H a =/ 0).

 12

When calculating the probability of the slope being significant the standard error

of the sample slope and the test statistic is found as:

s
SE b = √n
( S1x )

674.591 1
SE b = √10
( 3.028 )

S E b = 70.451

b−β
t= SE b

−570.295−0
t= 70.451

t =− 8.095

Using this test statistic as well as degrees of freedom value of 8 results in a

p-value of 2.005 × 10 −5 , or approximately 0, is found. Since the p-value is less than .05,

we can reject H 0 and say that there is a relationship between age and the annual price

of insurance.

Using the found LSRL, we can make predictions about the possible price of a

person at a given age between 16-25.

︿
y = 15268.668 − 570.295x
︿
y = 15268.668 − 570.295(22)
 13

︿
y = 15268.668 − 12, 546.49
︿
y = 2722.178

In this example, we can predict that at age 22, a person would have an annual

insurance premium of $2722.18. This is not the same as the average of the collected

prices for 22 year olds ($2323.96) possibly because of external factors affecting the

price, such as those discussed in the introduction.

Since the results are significant, creating a 95% confidence interval can help

show the relationship between age and premium price. Using a critical value of 1.860

from a calculator, a 95% confidence interval can be created:

b ± t * ( √sn )

− 570.295 ± 1.860( 674.591

√10
)

− 570.295 ± 396.783

(− 967.078,− 173.512)

The 95% confidence interval means that we can be 95% certain that the true

relationship between age and annual insurance premiums is between -967.078 and

-173.512. Since -570.295 is within this interval, it is a possible relationship between age

and annual price.

Another important pattern to note is the change seen in the average and

standard deviation of the annual prices with increasing age. The average annual price
 14

stays above $5000 until the age of 18, but is followed by a large fall to about $3500 at

19, and consistent decreases down to $1770.96 per year Standard deviation remains

above $1400 until the age of 18, then drops blow $1000 at 19, then decreases down to

$454.82. The largest drops in both of these values within this age group occur at the

age where students are leaving high school and becoming adults, which could be a

significant factor determining the annual price of insurance.

Old Age Group (35-85)

In the 35-85 age group there remains a strong, negative, linear relationship

between age and annual insurance premiums. A LSRL is calculated below:

√
n
∑ (y i −y) 2
i=1
Sy= n−1

√
6
∑ (y i −1602.330) 2
i=1
Sy= 5

√
263,924.498
Sy= 5

S y = 229.750
 15

√
n
∑ (xi −x) 2
i=1
Sx= n−1

√
6
∑ (xi −60) 2
i=1
Sx= 5

Sx=
√ 1750
5

S x = 18.708
rS y
b0 = Sx

(.611)(229.750)
b0 = (18.708)

b 0 = 7.504

b 1 = y − b 0x

b 1 = 1602.330 − (7.504)(60)

b 1 = 1152.09
︿
The resulting LSRL, y = 1152.09 + 7.504x , is pictured below. For the purposes of

remaining in the age group, the domain of the ages will be limited to 25 < x ≤ 85 .
 16

The correlation coefficient is as follows:

r = .661

r 2 = .373

37.3% of variation in insurance pricing is explained by age. For each additional

year of increased age, the predicted average cost of insurance increases by about

$7.504.

The residual plot shows a small

pattern in the data, but since the

data is limited a pattern is difficult

to define. Just as in the young age

group, this LSRL may not be

appropriate because of external

factors influencing the pricing of

car insurance. This is more apparent in the older age group because there are more

factors to consider when determining prices, such as the driver’s record. A test and

confidence interval will still be made with the LSRL in spite of these issues.

The null and alternate hypothesis will remain the same as in the young age

group, H 0 = 0 and H a =/ 0. The standard error of the sample slope and the test statistic

can be found in the same manner as previously used:

 17

s
SE b = √n
( S1x )

214.835 1
SE b = √6
( 18.708 )

S E b = 4.851

b−β
t= SE b

7.504−0
t= 4.851

t = 1.547

Using this test statistic and degrees of freedom value of 4, a p-value of .197 is obtained.

Since the p-value is greater than the significance level of .05, we can not reject H 0 and

do not say that there is a relationship between age and the annual price of insurance.

One reason that the results from the older age group are not significant is

because of the greater variety of factors influencing their price. A given individual’s

driving record, financial situation, or credit score might all have an unforeseen impact on

the price of insurance. Because of the high probability that the results are due to

random chance, it is not appropriate to make predictions with this model.

 18

Conclusion

Overall, the equation to predict the price of insurance based on age was more

accurate and useful for individuals between ages 16 and 25 while age did not act as a

good predictor for price on individuals in the older age bracket. This is largely due to the

increase in confounding variables affecting the data of older individuals, such as their

driving record and credit score.

Although the older age model is not the best for making a price prediction, The

model for younger drivers could be used by a car insurance company in order to give a

monthly premium price to individuals between ages 16 and 25.

As type I error is worse for a company providing car insurance, I used a smaller

significance level during my analysis. This was to minimize the likelihood that a car

company would use these results to make predictions when the results might have been

based on random chance. Although a type II error fails to provide a potentially usable

model for a car company, the potential of using an incorrect model poses a much

greater concern to a company than having to come up with a new model.

It is interesting to note that the results with the higher correlation coefficient

yielded a lower p-value. Further investigation into the relationship between correlation

coefficient and p-value may prove to be useful information.

 19

Works Cited

"Average Car Insurance Rates by Age." ​CarInsurance.com,​ QuinStreet Insurance Agency, 27

Nov. 2018, www.carinsurance.com/average-rates-by-age.aspx. Accessed 9 Dec. 2018.

Nau, Robert. "Mathematics of Simple Regression." ​Statistical Forecasting: Notes on Regression

and Time Series Analysis​, Duke University, people.duke.edu/~rnau/mathreg.htm.

Accessed 5 Nov. 2019.

Starnes, Daren S., et al. ​The Practice of Statistics.​ 5th ed., 2015.