Chapter 4 Displaying Quantitative Data

Transcription

1 Chapter 4 Displaying Quantitative Data Chapter 4 Displaying Quantitative Data Statistics in print. Answers will vary. 2. Not a histogram. Answers will vary. 3. Thinking about shape. a) The distribution of the number of speeding tickets each student in the senior class of a college has ever had is likely to be unimodal and skewed to the right. Most students will have very few speeding tickets (maybe 0 or 1), but a small percentage of students will likely have comparatively many (3 or more?) tickets. b) The distribution of player s scores at the U.S. Open Golf Tournament would most likely be unimodal and slightly skewed to the right. The best golf players in the game will likely have around the same average score, but some golfers might be off their game and score 15 strokes above the mean. (Remember that high scores are undesirable in the game of golf!) c) The weights of female babies in a particular hospital over the course of a year will likely have a distribution that is unimodal and symmetric. Most newborns have about the same weight, with some babies weighing more and less than this average. There may be slight skew to the left, since there seems to be a greater likelihood of premature birth (and low birth weight) than post-term birth (and high birth weight). d) The distribution of the length of the average hair on the heads of students in a large class would likely be bimodal and skewed to the right. The average hair length of the males would be at one mode, and the average hair length of the females would be at the other mode, since women typically have longer hair than men. The distribution would be skewed to the right, since it is not possible to have hair length less than zero, but it is possible to have a variety of lengths of longer hair. 4. More shapes. a) The distribution of the ages of people at a Little League game would likely be bimodal and skewed to the right. The average age of the players would be at one mode and the average age of the spectators (probably mostly parents) would be at the other mode. The distribution would be skewed to the right, since it is possible to have a greater variety of ages among the older people, while there is a natural left endpoint to the distribution at zero years of age. b) The distribution of the number of siblings of people in your class is likely to be unimodal and skewed to the right. Most people would have 0, 1, or 2 siblings, with some people having more siblings. c) The distribution of pulse rate of college-age males would likely be unimodal and symmetric. Most males pulse rates would be around the average pulse rate for college-age males, with some males having lower and higher pulse rates. d) The distribution of the number of times each face of a die shows in 100 tosses would likely be uniform, with around 16 or 17 occurrences of each face (assuming the die had six sides).

2 28 Part I Exploring and Understanding Data 5. Heart attack stays. The distribution of the length of hospital stays of female heart attack patients is skewed to the right, with stays ranging from 1 day to 36 days. The distribution is centered around 8 days, with the majority of the hospital stays lasting between 1 and 15 days. There are a relatively few hospital stays longer than 27 days. Many patients have a stay of only one day, possibly because the patient died. 6. s. The distribution of the number of s received from each student by a professor in a large introductory statistics class during an entire term is skewed to the right, with the number of s ranging from 1 to 21 s. The distribution is centered at about 2 s, with many students only sending 1 . There is one outlier in the distribution, a student who sent 21 s. The next highest number of s sent was only Sugar in cereals. a) The distribution of the sugar content of breakfast cereals is bimodal, with a cluster of cereals with sugar content around 10% sugar and another cluster of cereals around 48% sugar. The lower cluster shows a bit of skew to the right. Most cereals in the lower cluster have between 0% and 10% sugar. The upper cluster is symmetric, with center around 45% sugar. b) There are two different types of breakfast cereals, those for children and those for adults. The children s cereals are likely to have higher sugar contents, to make them taste better (to kids, anyway!). Adult cereals often advertise low sugar content. 8. Singers. a) The distribution of the heights of singers in the chorus is bimodal, with a mode at around 65 inches and another mode around 71 inches. No chorus member has height below 60 inches or above 76 inches. b) The two modes probably represent the mean heights of the male and female members of the chorus. 9. Vineyards. a) There is information displayed about 36 vineyards and it appears that 28 of the vineyards are smaller than 60 acres. That s around 78% of the vineyards. (75% would be a good estimate!) b) The distribution of the size of 36 Finger Lakes vineyards is skewed to the right. Most vineyards are smaller than 75 acres, with a few larger ones, from 90 to 160 acres. One vineyard was larger than all the rest, over 240 acres. The mode of the distribution is between 0 and 30 acres.

3 Chapter 4 Displaying Quantitative Data Run times. The distribution of runtimes is skewed to the right. The shortest runtime was around 28.5 minutes and the longest runtime was around 35.5 minutes. A typical run time was between 30 and 31 minutes, and the majority of runtimes were between 29 and 32 minutes. It is easier to run slightly slower than usual and end up with a longer runtime than it is to run slightly faster than usual and end up with a shorter runtime. This could account for the skew to the right seen in the distribution. 11. Gasoline. a) b) The distribution of gas prices is skewed to the right, centered around $2.10 per gallon, with most stations charging between $2.05 and $2.13. The lowest and highest prices were $2.03 and $2.27. c) There is a gap in the distribution of gasoline prices. There were no stations that charged between $2.19 and $ The Great One. a) Gasoline Prices in Ithaca Key: = $2.189/gal Wayne Gretzsky Games played per season Key: = 78 5 games b) The distribution of the number of games played by Wayne Gretzky is skewed to the left. c) Typically, Wayne Gretzky played about 80 games per season. The number of games played is tightly clustered in the upper 70s and low 80s. d) Two seasons are low outliers, when Gretzky played fewer than 50 games. He may have been injured during those seasons. Regardless of any possible reasons, these seasons were unusual compared to Gretzky s other seasons.

4 30 Part I Exploring and Understanding Data 13. Home runs. The distribution of the number of homeruns hit by Mark McGwire during the seasons is skewed to the right, with a typical number of homeruns per season in the 30s. With the exception of 3 seasons in which McGwire hit fewer than 10 homeruns, his total number of homeruns per season was between 22 and the maximum of Bird species. a) The results of the 1999 Laboratory of Ornithology Christmas Bird Count are displayed in the stem and leaf display at the right. This display uses split stems, to give the display a bit more definition. The lower stem contains leaves with digits 0,1,2,3,4 and the upper stem contains leaves with digits 5,6,7,8,9. b) The distribution of the number of birds spotted by participants in the 1999 Laboratory of Ornithology Christmas Bird Count is skewed right, with a center at around 160 birds. There are several high outliers, with two participants spotting 206 birds and another spotting 228. With the exception of these outliers, most participants saw between 152 and 186 birds. Christmas Bird Count Totals 1999 KEY: 18 6 = 186 species spotted. 15. Home runs, again. a) This is not a histogram. The horizontal axis should the number of home runs per year, split into bins of a convenient width. The vertical axis should show the frequency; that is, the number years in which McGwire hit a number of home runs within the interval of each bin. The display shown is a bar chart/time plot hybrid that simply displays the data table visually. It is of no use in describing the shape, center, spread, or unusual features of the distribution of home runs hit per year by McGwire. b) Mark McGwire s Home Runs Home runs per year

5 Chapter 4 Displaying Quantitative Data Return of the birds. a) This is not a histogram. The horizontal axis should split the number of counts from each site into bins. The vertical axis should show the number of sites in each bin. The given graph is nothing more than a bar chart, showing the bird count from each site as its own bar. It is of absolutely no use for describing the shape, center, spread, or unusual features of the distribution of bird counts. b) Number of sites Christmas Bird Count Number of species sighted 17. Horsepower. The distribution of horsepower of cars reviewed by Consumer Reports is nearly uniform. The lowest horsepower was 65 and the highest was 155. The center of the distribution was around 105 horsepower. 18. Population growth. The distribution of population growth in NE/MW states is unimodal, symmetric and tightly clustered around 5% growth. The distribution of population growth in S/W states is much more spread out, with most states having population growth between 5% and 30%. A typical state had about 15% growth. There were two outliers, Arizona and Nebraska, with 40% and 66% growth, respectively. In Generally, the growth rates in the S/W states were higher and more variable than the rates in the NE/MW states. 19. Hurricanes. a) A dotplot of the number of hurricanes each year from 1944 through 2000 is displayed. Each dot represents a year in which there were that many hurricanes. b) The distribution of the number of hurricanes per year is unimodal and skewed to the right, with center around 2 hurricanes per year. The number of hurricanes per year ranges from 0 to 7. There are no outliers. There may be a second mode at 5 hurricanes per year, but since there were only 4 years in which 5 hurricanes occurred, it is unlikely that this is anything other than natural variability. Consumer Reports Horsepower KEY: 11 5 = 115 horsepower OOOOO OOOOOOOOOOOOOO OOOOOOOOOOOOOOOOO OOOOOOOOOOOO OO OOOO OO O

6 32 Part I Exploring and Understanding Data 20. Hurricanes, again. The distribution of the number of hurricanes per year before 1970 is unimodal and skewed to the right. The center of the distribution is about 2 to 3 hurricanes per year. The number of hurricanes per year ranges from 0 to 7. After 1970, the distribution of the number of hurricanes per year is also unimodal and skewed right, with center around 1 or 2 hurricanes per year. The number of hurricanes per year ranges from 0 to 6. There may be a difference in the number of hurricanes per year before and after Before 1970, there may have been a slightly greater number of hurricanes in a typical year. 21. Acid rain. The distribution of the ph readings of water samples in Allegheny County, Penn. is bimodal. A roughly uniform cluster is centered around a ph of 4.4. This cluster ranges from ph of 4.1 to 4.9. Another smaller, tightly packed cluster is centered around a ph of 5.6. Two readings in the middle seem to belong to neither cluster. 22. Marijuana. The distribution of the percentage of 9 th graders in 20 Western European countries who have tried marijuana is unimodal and skewed to the right. Greece, at 2%, has the lowest percentage of 9 th graders who have tried marijuana. Scotland has the highest percentage, at 53%. A typical country might have a percentage of between 10% and 15%. Number of Hurricanes per Year OO 0 OOO OOO 1 OOOOOOOOOOO OOOOOOOOO 2 OOOOOOOO OOOOOO 3 OOOOOO OO 4 OO 5 OO O 6 O O 7 Acidity of Water Samples in Allegheny County, Penn. Percent of 9 th Graders in Western European Countries Who Have Tried Marijuana 23. Hospital stays. a) The histograms of male and female hospital stay durations would be easier to compare if the were constructed with the same scale, perhaps from 0 to 20 days

7 Chapter 4 Displaying Quantitative Data 33 b) The distribution of hospital stays for men is skewed to the right, with many men having very short stays of about 1 or 2 days. The distribution tapers off to a maximum stay of approximately 25 days. The distribution of hospital stays for women is skewed to the right, with a mode at approximately 5 days, and tapering off to a maximum stay of approximately 22 days. Typically, hospital stays for women are longer than those for men. c) The peak in the distribution of women s hospital stays can be explained by childbirth. This time in the hospital increases the length of a typical stay for women, and not for men. 24. Deaths. According to the National Vital Statistics report in 1999, there were several key differences between the distributions of age at death for Black Americans and White Americans. The distribution of age at death for Black Americans was skewed left, with a center at approximately 65 to 75 years of age. There was a cluster of ages at death corresponding to the very young. The distribution of age at death for White Americans was also skewed to the left, although to a greater extent than the distribution of age at death for Black Americans. White Americans had a center at approximately 75 to 85 years old at death, roughly 10 years higher than Black Americans. Additionally, the cluster of ages at death corresponding to the very young was much smaller for White Americans than for Blacks, probably indicating a higher infant mortality rate for Black Americans. 25. Final grades. The width of the bars is much too wide to be of much use. The distribution of grades is skewed to the left, but not much more information can be gathered. 26. Cities. a) The distribution of cost of living in 25 international cities is unimodal and skewed to the right. The distribution is centered around $100, and spread out, with values ranging from $60 to $180. b) The most expensive city included here, Tokyo, does not appear to be an outlier. It seems to be the end of a long tail. 27. Final grades revisited. a) This display has a bar width that is much too narrow. As it is, the histogram is only slightly more useful than a list of scores. It does little to summarize the distribution of final exam scores. b) The distribution of test scores is skewed to the left, with center at approximately 170 points. There are several low outliers below 100 points, but other than that, the distribution of scores is fairly tightly clustered. 28. Cities revisited. a) The distribution of cost of living in 25 international cities now appears bimodal, with many cities costing just under $100 and another smaller cluster around $140. b) Either answer is OK. You have the right to decide that Tokyo is an outlier, or that it is not sufficiently extreme. Certainly, this histogram makes Tokyo appear more extreme than the histogram in Exercise 26.

8 34 Part I Exploring and Understanding Data 29. Zip codes. Even though zipcodes are numbers, they are not quantitative in nature. Zipcodes are categories. A histogram is not an appropriate display for categorical data. The histogram the Holes R Us staff member displayed doesn t take into account that some 5-digit numbers do not correspond to zipcodes or that zipcodes falling into the same classes may not even represent similar cities or towns. The employee could design a better display by constructing a bar chart that groups together zipcodes representing areas with similar demographics and geographic locations. 30. CEO data revisited. a) First of all, it must be noted that industry codes are categorical, so the use of a histogram as a display is inappropriate. Strangely enough, the investment analyst made even more mistakes! This display is really just a poorly constructed bar chart (of sorts). There are gaps in the display because all of the industry codes are integers and the widths of the bars are all less than 1. With 5 bars between 0 and 3.75, we can find the width to be So, for example, there appears to be a gap between 2.25 and 3, simply because there are no integers between 2.25 and 3 (remember, the upper boundary is not inclusive). The gaps aren t really there, but a poor choice of scale makes them appear. b) This question doesn t really make any sense. Unimodal is a vocabulary word specific to describing distributions of quantitative data. As mentioned before, the industry codes are categorical. c) A histogram can never be used to summarize categorical data. The analyst would be better off displaying the data in a bar chart, with relative heights of the bars representing the number of CEOs involved in each industry, or a pie chart displaying the percentage of CEOs involved in each industry. 31. Productivity study. The productivity graph is useless without a horizontal scale indicating the time period over which the productivity increased. Also, we don t know the units in which productivity is measured. 32. Productivity revisited. The display of productivity and wages display has no scale on either axis and no units are indicated. For the vertical axis, it is unlikely that wages and productivity can be measured meaningfully in the same units, so the two are almost certainly incomparable. Also, we don t know the time scale (on the horizontal axis) over which productivity and wages were measured. In fact, given the problems on the vertical axis, it is not even apparent that the horizontal axis has comparable time periods for wages and productivity.

9 Chapter 4 Displaying Quantitative Data Law enforcement. a) The histograms at the right show the number of assaults per 1000 officers and number of killed or injured per 1000 officers, for eleven federal agencies with officers authorized to carry firearms and make arrests. b) The distribution of assault rates for these federal agencies is roughly symmetric with two high outliers. The center of the Assaults (per 1000) distribution is between 5 and 10 assaults per 1000 officers and, with the exception of two agencies, BATF and National Park Service, with over 30 assaults per 100 officers, the distribution is tightly clustered. The distribution of killed and injured rates for these eleven law enforcement agencies is very tightly clustered with almost all of the agencies reporting rates of between 0 and 5 officers per 1000 killed and injured. Customs Service, with a rate of 5.1 officers per 1000, is essentially part of this group, as well. There is one high outlier, the National Park Service, with a rate of 15 officers per 1000 killed and injured. c) The National Park Service is a high outlier for assaults and killed-injured, with rates of 38.7 officers per 1000 assaulted and 15 officers per 1000 killed or injured. The BATF is a high outlier for assaults, with 31.1 officers per 1000 assaulted. 34. Cholesterol. The distribution of cholesterol levels for smokers is unimodal and skewed slightly to the right, with a mode around 210. Cholesterol levels vary from approximately 140 to 350, but are generally clustered between 200 and 300. There is one low cholesterol level and one high cholesterol level, but these don t seem to depart from the overall pattern. The distribution of cholesterol levels for nonsmokers is unimodal and roughly symmetric, with a center around 240. Cholesterol levels vary from approximately 120 to 340, and seem spread out. There is one low cholesterol level, but not unusually low. In general, the cholesterol levels of smokers seem to be slightly lower than the cholesterol levels of non-smokers. Additionally, the cholesterol levels of smokers appear more consistent than cholesterol levels of non-smokers. Killed Injured (per 1000)

10 36 Part I Exploring and Understanding Data 35. MPG. a) A back-to-back stemplot of these data is shown at the right. A plot with with tens and units digits for stems and tenths for leaves would have been quite long, but still useable. A plot with tens as stems and rounded units as leaves would have been too compact. This plot has tens as the stems, but the stems are split 5 ways. The uppermost 2 stem displays 29 and 28, the next 2 stem displays 27 and 26, and so on. The key indicates the rounding used, as well as the accuracy of the original data. In this case, the mileages were given to the nearest tenth. US Cars Other Cars KEY: 2 5 = mpg b) In general, the Other cars got better gas mileage than the US Cars, although both distributions were highly variable. The distribution of US cars was bimodal and skewed to the right, with many cars getting mileages in the high teens and low twenties, and another group of cars whose mileages were in the high twenties and low thirties. Two high outliers had mileages of 34 miles per gallon. The distribution of Other cars, in contrast, was bimodal and skewed to the left. Most cars had mileages in the high twenties and thirties, with a small group of cars whose mileages were in the low twenties. Two low outliers had mileages of 16 and 17 miles per gallon. 36. Baseball. American National a) The back-to-back stemplot shown at the right has split stems to show the distribution in a bit more detail than a stemplot with single stems. b) The distribution of number of runs per game in stadiums in the American League is unimodal and slightly skewed to the right, clustered predominantly in the interval of 9 to 10 runs per game. In the National League, the number of runs scored per game is distributed symmetrically and is possibly bimodal, with clusters in the high 9s and low 10s and also in the low 8s. There are two high outliers of 11.6 and 14 runs per game. The number of runs scored per game is generally higher and more consistent in the American League. KEY: 10 3 = 10.3 runs per game c) The 14 runs per game scored at Coors Field is an outlier in the National League data for the first half of the 2001 season. There appear to be more runs per game scored there than in other Major League Stadiums.

11 Chapter 4 Displaying Quantitative Data Nuclear power. a) The stemplot at the right shows the distribution of construction costs, measured in $1000 per mw of 12 nuclear power generators. b) The distribution of nuclear power plant construction costs is skewed left, and has several modes, with gaps in between. Several plants had costs near $80,000 per mw, another few had costs near $60,000 per mw and several had costs near $30,000 per mw. c) The timeplot at the right shows the change in nuclear plant construction costs over time. Nuclear Plant Construction Costs Construction Cost ($1000/mW) Key: = $ per mw d) From the timeplot, it is apparent that the nuclear plant construction costs generally increased over time. 38. Drunk driving. a) The stemplot (near right) shows the distribution of drunk driving deaths. b) The timeplot (far right) shows the change in drunk driving deaths over time. KEY: 22 4 = 22,400 deaths Drunk Driving Deaths Drunk Driving Deaths c) The distribution of the number of drunk driving deaths is bimodal, with a cluster between 22 and 25 thousand deaths and another cluster between 16 and 17 thousand deaths. The timeplot shows that this corresponds to a rapid decrease in the drunk driving deaths in the early nineties. Previously, the number of deaths was high, then decreased dramatically.

12 38 Part I Exploring and Understanding Data 39. Assets. a) The distribution of assets of 79 companies chosen from the Forbes list of the nation s top corporations is skewed so heavily to the right that the vast majority of companies have assets represented in the first bar of the histogram, 0 to 10 billion dollars. This makes meaningful discussion of center and spread impossible. b) Re-expressing these data by, for example, logs or square roots might help make the distribution more nearly symmetric. Then a meaningful discussion of center might be possible. 40. Music Library. a) The distribution of the number of songs students had in their digital libraries is extremely skewed to the right. That makes it difficult to determine a center. The typical number of songs in a library is probably in the first bar of the histogram. b) Re-expressing these data by, for example, logs or square roots might help make the distribution more nearly symmetric. Then a meaningful discussion of center might be possible. 41. Assets again. a) The distribution of logarithm of assets is preferable, because it is roughly unimodal and symmetric. The distribution of the square root of assets is still skewed right, with outliers. b) If Assets = 50, then the companies assets are approximately 50 2 = 2500 million dollars. c) If log( Assets ) = 3, then the companies assets are approximately 10 3 = 1000 million dollars. 42. Rainmakers. a) Since one acre-foot is about 320,000 gallons, these numbers are more manageable than gallons. b) The distribution of rainfall from 26 clouds seeded with silver iodide is skewed heavily to the right, with the vast majority of clouds producing less than 500 acrefeet of rain. Several clouds produced more, with a maximum of 2745 acre-feet.

13 Chapter 4 Displaying Quantitative Data 39 c) The distribution of log (base 10) of rainfall is much more symmetric than the distribution of rainfall. We can see that the center of the distribution is around log 2 log 2.5 acre-feet. d) Since the reexpressed scale is measured in log (acrefeet), we need to raise 10 to the power of the number on our scale to convert back to acre feet. For example, if a cloud in the new scale has a log (rainfall) of 2.3, we convert back to rainfall as follows: log( rainfall) = rainfall = 10 rainfall = Log(Rainfall) The cloud produced acre-feet of rain.