We have spent most of this chapter analyzing models described by graphs or equations. To create a model, however, we often start with a quantity of data. Choosing an appropriate function for a model is a complicated process. In this section, we consider only linear models and explore methods for fitting a linear function to a collection of data points. First, we fit a line through two data points.
If we already know that two variables are related by a linear function, we can find a formula from just two data points. For example, variables that increase or decrease at a constant rate can be described by linear functions.
In 1993, Americans drank 188.6 million cases of wine. Wine consumption increased at a constant rate over the next decade, and we drank 258.3 million cases of wine in 2003. (Source: Los Angeles Times, Adams Beverage Group)
Find a formula for wine consumption, \(W\text{,}\) in millions of cases, as a linear function of time, \(t\text{,}\) in years since 1990.
We have two data points of the form \((t, W)\text{,}\) namely \((3, 188.6)\) and \((13, 258.3)\text{.}\) We use the point-slope formula to fit a line through these two points. First, we compute the slope.
In 1991, there were 64.6 burglaries per 1000 households in the United States. The number of burglaries reported annually declined at a roughly constant rate over the next decade, and in 2001 there were 28.7 burglaries per 1000 households. (Source: U.S. Department of Justice)
Find a function for the number of burglaries, \(B\text{,}\) as a function of time, \(t\text{,}\) in years, since 1990.
Because \(t=0\) corresponds to 1990, we have the two points \((1,64.6)\) and \((11,28.7)\) to compute the slope \(m=-3.59\text{.}\) Using the point-slope formula with either point and then simplifying, we find that \(B = 68.19 - 3.59t\)
Empirical data points in a linear relation may not lie exactly on a line. There are many factors that can affect experimental data, including measurement error, the influence of environmental conditions, and the presence of related variable quantities.
A consumer group wants to test the gas mileage of a new model SUV. They test-drive six vehicles under similar conditions and record the distance each drove on various amounts of gasoline.
No, the data are not strictly linear. If we compute the slopes between successive data points, the values are not constant. We can see from an accurate plot of the data, shown below, that the points lie close to, but not precisely on, a straight line.
We would like to draw a line that comes as close as possible to all the data points, even though it may not pass precisely through any of them. In particular, we try to adjust the line so that we have the same number of data points above the line and below the line. One possible solution is shown above.
To compute the slope of the our estimated line, we first choose two points on the line. Our line appears to pass through one of the data points,\((8.8, 139.6)\text{.}\) We look for a second point on the line whose coordinates are easy to read, perhaps \((6.5,100)\text{.}\) The slope is
\begin{equation*}
m = \frac{139.6 - 100}{8.8 - 6.5}= 17.2\text{ miles per gallon}
\end{equation*}
According to our data, the SUV gets about 17.2 miles to the gallon.
To find the slope of your estimated line, be sure to choose points on the line; do not choose any of the data points (unless they happen to lie on your line).
The graph in ExampleΒ 1.146 is called a scatterplot. The points on a scatterplot may or may not show some sort of pattern. Consider the three plots shown below.
In figure (a), the data points resemble a cloud of gnats; there is no apparent pattern to their locations.
If the data in a scatterplot are roughly linear, we can estimate the location of an imaginary line of best fit that passes as close as possible to the data points. We can then use this line to make predictions about the data.
One measure of a personβs physical fitness is the body mass index, or BMI. Your BMI is the ratio of your weight in kilograms to the square of your height in centimeters. Thus, thinner people have lower BMI scores, and fatter people have higher scores. The Centers for Disease Control considers a BMI between 18.5 and 24.9 to be healthy.
The points on the scatterplot below show the BMI of Miss America from 1921 to 1991. From the data in the scatterplot, can we see a trend in Americansβ ideal of female beauty?
We draw a line that fits the data points as best we can, as shown below. (Note that we have set \(t = 0\) in 1920 on this graph.) We try to end up with roughly equal numbers of data points above and below our line.
We see that when \(t = 60\) on this line, the \(y\)-value is approximately 18.3. We therefore estimate that Miss America 1980 had a BMI of 18.3. (Her actual BMI was 17.85.)
Human brains consume a large amount of energy, about 16 times as much as muscle tissue per unit weight. In fact, brain metabolism accounts for about 25% of an adult humanβs energy needs, as compared to about 5% for other mammals. As hominid species evolved, their brains required larger and larger amounts of energy, as shown in the figure. (Source: Scientific American, December 2002.)
The process of predicting an output value based on a straight line that fits the data is called linear regression, and the line itself is called the regression line. The equation of the regression line is usually used (instead of a graph) to predict values.
We first calculate the slope by choosing two points on the regression line. The points we choose are not necessarily any of the original data points; instead they should be points on the regression line itself. The line appears to pass through the points \((17, 20)\) and \((67, 18)\text{.}\) The slope of the line is then
\begin{equation*}
m = \frac{18 - 20}{67 - 17}\approx -0.04
\end{equation*}
Now we use the point-slope formula to find the equation of the line. (If you would like to review the point-slope formula, see SectionΒ 1.5.) We substitute \(m = -0.04\) and use either of the two points for \((x_1, y_1)\text{;}\) we will choose \((17, 20)\text{.}\) The equation of the regression line is
\begin{equation*}
\begin{aligned}[t]
y \amp = y_1 + m(x - x_1)\\
y \amp = 20-0.04(x-17) \amp \amp \blert{\text{Simplify.}}\\
y \amp = 20.68 - 0.04t\\
\end{aligned}
\end{equation*}
The number of manatees killed by watercraft in Florida waters has been increasing since 1975. Data are given at 5-year intervals in the table. (Source: Florida Fish and Wildlife Conservation Commission)
Using a regression line to estimate values between known data points is called interpolation. Making predictions beyond the range of known data is called extrapolation.
Our model predicts that Miss America 2001 had a BMI of 17.44. In fact, her BMI was 20.25. By the late 1990s, public concern over the self-image of young women had led to a reversal of the trend toward ever-thinner role models.
ExampleΒ 1.157b illustrates an important fact about extrapolation: If we try to extrapolate too far, we may get unreasonable results. For example, if we use our model to predict the BMI of Miss America 2520 (when \(t = \alert{600}\)), we get
\begin{equation*}
y = 20.68 - 0.04(\alert{600}) =-3.32
\end{equation*}
Even if the Miss America pageant is still operating in 600 years, the winner cannot have a negative BMI. Our linear model provides a fair approximation for 1920β1990, but if we try to extrapolate too far beyond the known data, the model may no longer apply.
We can also use interpolation and extrapolation to make estimates for nonlinear functions. Sometimes a variable relationship is not linear, but a portion of its graph can be approximated by a line.
For example, the graph at right shows a childβs height each month. The graph is not linear because her rate of growth is not constant; her growth slows down as she approaches her adult height. However, over a short time interval the graph is close to a line, and that line can be used to approximate the coordinates of points on the curve.
Estimating a line of best fit is a subjective process. Rather than base their estimates on such a line, statisticians often use the least squares regression line.
This regression line minimizes the sum of the squares of all the vertical distances between the data points and the corresponding points on the line, as shown at left. Many calculators are programmed to find the least squares regression line, using an algorithm that depends only on the data, not on the appearance of the graph.
Technology1.161Using Technology for Linear Regression.
You can use a graphing calculator to make a scatterplot, find a regression line, and graph the regression line with the data points. On the TI-84 calculator, we use the statistics mode, which you can access by pressing STAT. You will see a display that looks like figure (a) below. Choose \(1\) to \(Edit\) (enter or alter) data.
If there are data in column \(L_1\) or \(L_2\text{,}\) clear them out: Use the β key to select \(L_1\text{,}\) press CLEAR, then do the same for \(L_2\text{.}\)
Enter the \(x\)-coordinates of the data points in the \(L_1\) column and enter the \(y\)-coordinates in the \(L_2\) column, as shown in figure (a) below.
When you are through with the scatterplot, press Y=βENTER to turn off the \(Stat Plot\text{.}\) If you neglect to do this, the calculator will continue to show the scatterplot even after you ask it to plot a new equation.
It cost a bicycle company $\(9000\) to make \(40\) touring bikes in its first month of operation and $\(15,000\) to make \(125\) bikes during its second month. Express the companyβs monthly production cost, \(C\text{,}\) in terms of the number, \(x\text{,}\) of bikes it makes.
Flying lessons cost $\(645\) for an \(8\)-hour course and $\(1425\) for a \(20\)-hour course. Both prices include a fixed insurance fee. Express the cost, \(C\text{,}\) of flying lessons in terms of the length, \(h\text{,}\) of the course in hours.
Under ideal conditions, Andreaβs Porsche can travel \(312\) miles on a full tank (\(12\) gallons of gasoline) and \(130\) miles on \(5\) gallons. Express the distance, \(d\text{,}\) Andrea can drive in terms of the amount of gasoline, \(g\text{,}\) she buys.
On an international flight, a passenger may check two bags each weighing \(70\) kilograms, or \(154\) pounds, and one carry-on bag weighing \(50\) kilograms, or \(110\) pounds. Express the weight, \(p\text{,}\) of a bag in pounds in terms of its weight, \(k\text{,}\) in kilograms.
A radio station in Detroit, Michigan, reports the high and low temperatures in the Detroit/Windsor area as \(59\degree\)F and \(23\degree\)F, respectively. A station in Windsor, Ontario, reports the same temperatures as \(15\degree\)C and \(-5\degree\)C. Express the Fahrenheit temperature, \(F\text{,}\) in terms of the Celsius temperature, \(C\text{.}\)
Ms. Randolph bought a used car in 2000. In 2002, the car was worth \($9000\text{,}\) and in 2005 it was valued at \($4500\text{.}\) Express the value, \(V\) , of Ms. Randolphβs car in terms of the number of years, \(t\text{,}\) she has owned it.
Use the equation to predict the number of cups of cocoa that will be sold when the temperature is \(10\degree\)C and when the temperature is \(24\degree\)C. Are these predictions reasonable?
With Americansβ increased use of faxes, pagers, and cell phones, new area codes are being created at a steady rate. The table shows the number of area codes in the United States each year. (Source: USA Today, NeuStar, Inc.)
The number of mobile homes in the United States has been increasing since 1960. The data in the table are given in millions of mobile homes. (Source: USA Today, U.S. Census Bureau)
Teenage birth rates in the United States declined from 1991 to 2000. The table shows the number of births per 1000 women in selected years. (Source: U.S. National Health Statistics)
Life expectancy in the United States has been rising since the nineteenth century. The table shows the U.S. life expectancy in selected years. (Source: http://www.infoplease.com)
"The earnings gap between high-school and college graduates continues to widen, the Census Bureau says. On average, college graduates now earn just over $\(51,000\) a year, almost twice as much as high-school graduates. And those with no high-school diploma have actually seen their earnings drop in recent years." The table shows the unemployment rate and the median weekly earnings for employees with different levels of education. (Source: Morning Edition, National Public Radio, March 28, 2005)
Do you think this model is useful for extrapolation or interpolation? For example, what weekly earnings does the model predict for someone with 15 years of education? For 25 years? Do you think these predictions are valid? Why or why not?
The table shows the birth rate (in births per woman) and the female literacy rate (as a percent of the adult female population) in a number of nations. (Source: UNESCO, The World Fact Book, EarthTrends)
The table shows the amount of carbon released into the atmosphere annually from burning fossil fuels, in billions of tons, at 5-year intervals from 1950 to 1995. (Source: www.worldwatch.org)
High-frequency radiation is harmful to living things because it can cause changes in their genetic material. The data below, collected by C. P. Oliver in 1930, show the frequency of genetic transmutations induced in fruit flies by doses of X-rays, measured in roentgens. (Source: C. P. Oliver, 1930)
Bracken, a type of fern, is one of the most successful plants in the world, growing on every continent except Antarctica. New plants, which are genetic clones of the original, spring from a network of underground stems, or rhizomes, to form a large circular colony. The graph shows the diameters of various colonies plotted against their age. (Source: Chapman et al.,1992)
The European sedge warbler can sing several different songs consisting of trills, whistles, and buzzes. Male warblers who sing the largest number of songs are the first to acquire mates in the spring. The data below show the number of different songs sung by several male warblers and the day on which they acquired mates, where day 1 is April 20. (Source: Krebs and Davies, 1993)
Plot the data points, with number of songs on the horizontal axis. A regression line for the data is \(y = -0.85x + 53\text{.}\) Graph this line on the same axes with the data.
One of the factors that determines the strength of a muscle is its cross-sectional area. The data below show the cross-sectional area of the arm flexor muscle for several men and women, and their strength, measured by the maximum force they exerted against a resistance. (Source: Davis, Kimmet, Autry, 1986)
Astronomers use a numerical scale called magnitude to measure the brightness of a star, with brighter stars assigned smaller magnitudes. When we view a star from Earth, dust in the air absorbs some of the light, making the star appear fainter than it really is. Thus, the observed magnitude of a star, \(m\text{,}\) depends on the distance its light rays must travel through the Earthβs atmosphere. The observed magnitude is given by
\begin{equation*}
m = m_0 + kx
\end{equation*}
where \(m_0\) is the actual magnitude of the star outside the atmosphere, \(x\) is the air mass (a measure of the distance through the atmosphere), and \(k\) is a constant called the extinction coefficient. To calculate \(m_0\text{,}\) astronomers observe the same object several times during the night at different positions in the sky, and hence for different values of \(x\text{.}\) Here are data from such observations. (Source: Karttunen et al., 1987)
Six students are trying to identify an unknown chemical compound by heating the substance and measuring the density of the gas that evaporates. (Density \(=\) mass/volume.) The students record the mass lost by the solid substance and the volume of the gas that evaporated from it. They know that the mass lost by the solid must be the same as the mass of the gas that evaporated. (Source: Hunt and Sykes, 1984)
The formulas for many chemical compounds involve ratios of small integers. For example, the formula for water, H\(_2\)0, means that two atoms of hydrogen combine with one atom of oxygen to make one water molecule. Similarly, magnesium and oxygen combine to produce magnesium oxide. In this problem, we will discover the chemical formula for magnesium oxide. (Source: Hunt and Sykes, 1984)
Twenty-four grams of magnesium contain the same number of atoms as sixteen grams of oxygen. Complete the table showing the amount of oxygen needed if the formula for magnesium oxide is \(\text{MgO}\text{,}\)\(\text{Mg}_2\text{O}\text{,}\) or \(\text{MgO}_2\text{.}\)
Graph three lines on the same axes to represent the three possibilities, with grams of magnesium on the horizontal axis and grams of oxygen on the vertical axis.
Newborn blue whales are about \(24\) feet long and weigh \(3\) tons. The young whale nurses for \(7\) months, at which time it is \(53\) feet long. Estimate the length of a \(1\)-year-old blue whale.
A car starts from a standstill and accelerates to a speed of \(60\) miles per hour in \(6\) seconds. Estimate the carβs speed \(2\) seconds after it began to accelerate.
A truck on a slippery road is moving at \(24\) feet per second when the driver steps on the brakes. The truck needs \(3\) seconds to come to a stop. Estimate the truckβs speed \(2\) seconds after the brakes were applied.
The temperature of an automobile engine is \(9\degree\) Celsius when the engine is started and \(51\degree\)C seven minutes later. Use a linear model to predict the engine temperature for both \(2\) minutes and \(2\) hours after it started. Are your predictions reasonable?
The temperature in Death Valley is \(95\degree\) Fahrenheit at 5 a.m. and rises to \(110\degree\) Fahrenheit by noon. Use a linear model to predict the temperature at 2 p.m. and at midnight. Are your predictions reasonable?
Ben weighed \(8\) pounds at birth and \(20\) pounds at age \(1\) year. How much will he weigh at age \(10\) if his weight increases at a constant rate?
The elephant at the City Zoo becomes ill and loses weight. She weighed \(10,012\) pounds when healthy and only \(9641\) pounds a week later. Predict her weight after \(10\) days of illness.
Birdsβ nests are always in danger from predators. If there are other nests close by, the chances of predators finding the nest increase. The table shows the probability of a nest being found by predators and the distance to the nearest neighboring nest. (Source: Perrins, 1979)
A trained cyclist pedals faster as he increases his cycling speed, even with a multiple-gear bicycle. The table shows the pedal frequency, \(p\) (in revolutions per minute), and the cycling speed, \(c\) (in kilometers per hour), of one cyclist. (Source: Pugh, 1974)
In this problem we will calculate the efficiency of swimming as a means of locomotion. A swimmer generates power to maintain a constant speed in the water. If she must swim against an opposing force, the power increases. The following table shows the power expended by a swimmer while working against different amounts of force. (A positive force opposes the swimmer, and a negative force helps her.) (Source: diPrampero et al., 1974, and Alexander, 1992)
Plot the data on the grid, or use the StatPlot feature on your calculator. Use your calculator to find the least squares regression line. Graph the regression line on top of the data.
The swimmerβs mechanical power (or rate of work) is computed by multiplying her speed times the force needed to tow her at rest. Use your answer to part (d) to calculate the mechanical power she generates by swimming at \(0.4\) meters per second.
The ratio of mechanical power to metabolic power is a measure of the swimmerβs efficiency. Compute the efficiency of the swimmer when there is no external force opposing or helping her.
In this problem, we calculate the amount of energy generated by a cyclist. An athlete uses oxygen slowly when resting but more quickly during physical exertion. In an experiment, several trained cyclists took turns pedaling on a bicycle ergometer, which measures their work rate. The table shows the work rate of the cyclists, in watts, measured against their oxygen intake, in liters per minute. (Source: Pugh, 1974)
Plot the data on the grid, or use the StatPlot feature on your calculator. Use your calculator to find the least squares regression line. Graph the regression line on top of the data.
What is the slope of the regression line? The slope represents the amount of power, in watts, generated by a cyclist for each liter of oxygen consumed per minute. How many watts of power does a cyclist generate from each liter of oxygen?