Objective 1 Define the nll, ‘and alternate hypotheses The Null Hypothesis and the Alternate Hypothesis Air pollution has become a serious health problem in many cities. One of the forms of air pollution that healt officals are most concerned about is particulate matter (PM), which refers to fine particles that can be trapped in the lungs, increasing the risk of respiratory disease. Some of the PM in the atmosphere comes from car exhaust, so one important way to reduce PM polation isto design automobile engines that produce less PM. The following ‘example will show how hypothesis testing can play a part in this effort. ‘A study published in the Journal of the Air and Waste Management Association reported that the mean amount of PM produced by cars and fight trucks in an urban seiting is 35 milligrams of PM per mile of travel. Suppose that a new engine design is proposed that is intended to reduce this level. Now there are two possibilities: either the ‘new design will reduce the Level, ort will not. These possibilities are called hypotheses. To 1. The null hypashesis says that the new design will not reduce the level, so the mean for the new engines will bey = 35, 2. The alternate hypothesis says that the new design will reduce the level, so 4 < 35. ‘In general, the null hypothesis says that a parameters equal toacertain value, while the alternate hypothesis say tha the parameter differ from this value. Often the nul hypothesis is a statement of no change or no difference, while the alternate hypothesis states that a change or difference has occurred.Section 9.1 Basic Principles of Hypothess Testing 433 abot a parameter tes that the parameter is equal or example, Hg: = 35. The null hypothesis is denoted Mo alternate hypothesks about apa-areter states thatthe value of the paramtes ‘om the value specified by tne mill hypothesis. The alternate hypothesis is eno #, ‘There ae three types of aerate hypothesis, which we now define. + A lefietatedsliernsie iyposhesis sts thatthe parameter is less than the value speciied by tne oul hypothesis or example, Hy: 4 < 35. *+ Arright-taled aerate bypothess states thatthe parameter i greater than the value specified by the null hypothesis, for example, Hy: > 35. + A totalled aerate hypothesis states thatthe parameter is not equal tothe value specified by the null hypothesis, for example, ff: j 3. Left-tiled and righ-tailed hypotheses: led ied hypotheses SS Example 9.1 State the null and alternate hypotheses Boxes ofacertain kindof cereal ar labeled as contining 20 ounces. An inspector thinks that the mean weight may be les than this, State the appropriate null and alternate hypotheses, Solution ‘Toe null hypothesis says that there so difference, so the ml hypothesis i H: w= 20. ‘The inspector thinks thatthe mean weight may be less than 20, s the alternate hypothesis is Hy w < 20. Example 9.2 State the null and alternate hypotheses Las year, the mean monthly rent for an apartment ina certain city was $800. A ral estate agent believes thatthe mean rents higher this year, tat the appropriate nul and alternate hypotheses. Solution “The null hypobesis ays that there is no change, othe nll hypothesis Hy real estale agent wants to know whether the mean is higher, so the alternate hypothesis is Hy p> 800. Example 9.3 State the null and alternate hypotheses Scores on a standardized test have a mean of 70. Some modifications are made tothe test, ‘and an educator believes thatthe mean may have changed. State the appropriate null and alternate hypotheses. Solution ‘The null hypothesis says that there is no change, so the null hypothesis is Mg: = 70. The educator wants to know whether the mean has changed, without specifying whether it has increased or decreased. Therefore, the altermate hypothesis is Hy: 4 # 70.Objective 2 Descrive the reasoning used in hypothesis testing A hypothesis test is like a trial The purpose ofa hypothe:s test isto determine how piausible the aul hypothesis is. The ides behind hypothesis testing is the same as the idea behind’ a criminal wil, At the start i assumed to be innocent. Then the evidence is presented. Ifthe ce strony indicates that the defendant is guilty, we abandon the assumption of in- ce and find the defendant guilty. ina hypothesis cst the aull hypothesis plays the role ofthe defendant At the start ofa hypothesis test, we assume that the null hypothesis is true. Then we look atthe evidence, which comes from data that have been collected. Ifthe data strongly indicate thatthe null hypothesis is false, we abandon our assumption that it is ue and believe the alternate hypothesis instead. This is referred to as rejecting the null hypothesis. SUMMARY *+ We begin a hypothesis test by assuming the null hypothesis to be rue. * Ifthe data provide strong evidence against the null hypothesis, we reject it, and believe the alternate hypothesis. al. the defenda The Reasoning Used in Hypothesis Testing ‘We reject the null hypothesis when the data provide trong evidence against. The evidence is im the form of a statistic calculated from the data. In many cases, we make a decision by ‘computing the difference between this statistic and te parameter value specified by the null hypothesis. When tis difference is sufiiently large, we reject Ho. For example, imagine that an exam has been given 0 a large class. Someone claims that the mean score onthe exam was 80. This isthe nll hypothesis: Hw = 80. Assume the alternate hypothesis is H,: 4 # 80. Now you sample some students and find thatthe sample mean exam score is = = 78. This is not exactly equal to the null mean jt = 80. However, we expect the sample mean to differ somewhat from the population mean, just by chance, Because 78 is fairly close to 80, itis plausible thatthe nul hypothesis is true: the population mean is « = 80, Now imagine thatthe sample mean had been X = 50.’This value is far from the null mean of 80, and itis unlikely tha diffrence this large is du just to chance. Because 50 is far from 80, it does not seem plausible thatthe population mean is 80. Therefore, we would probably reject Hy. See Figures 9.1 and 9.2 on page 435, ‘We see thatthe farther the test statistic is from the parameter value specified by Hp the tess likely the difference is due to chance, and te less plausible Hy becomes. The question then is: How big does the diference have tobe before we reject Ho? To answer this question, ‘we need methods that enable us o calculate just how plausible His. Hypothesis tests pro these methods taking into account things such as the size ofthe sarmpe and the amount ‘of spread in the distribution. Inthe remaining sections of this chapter, we will Iearn ways tb calculate a value that tells us how plausible Hy i, so we can decide whether to reject itSection 9.1. Resi Principles of Hypothesis Testing 435, 2 % Figure 9.1 The nt hypothesis isi: y = 90 Te i= 78 Thesamplemeanincos mag ae ype, 0 9 Ne sample ncan Figure 9.2 Te mall hypothesis is: y= 0, The spe mean Slate vane sic! yell n=. Thesanple mean ifarrom eal pected etal Ld mt rejee Ho. rs 0 we wold ject Objective 3 sie Stating Conclusions 1 the nll typothess is rete, we conclge that Hs tue, We can sate this conclusion by expressing yn words. We shld nt imply say “we eject the wll hypothe a eee State a conclusion when the null hypothesis is rejected Boxes ofa cena kind of eral are Inbeed as containing 20 ounces, An inspector thinks ‘hat the mean weight may be es than ths, so he performs ates of Hy: y= 20 vers My: 4 < 20, He rejects the null hypotesis. State an appropriate conclusion, Solution Because the null hypothesis is rejected, we conclude that the aterate hypothesis sue ‘We express te alierate hypothesis in words "We conclude thatthe mean weight of ees! boxes is ess than 20 ounces.” ener ke a IF the null hypothesis is rejected, the conclusion is straightforward: We conclude that EXPLAIN IT AGAIN the nul hype i false andthe lterate hypetbess is tue. However, ifthe ml hype The conclusion of hypothesis «ess not ected, we do nr conclude that the nal hypothesis is true. In our formation, testis the the verdict of une ‘the null hypothesis sys that parameter, such a is equal to a certain value Now we Notrejecting Hp isthe ay can never be suf that a parameter is exactly equal o& particular value, Therefore, we con Verdict of not uit Arnot guy never be sure ha the nll hypothesis is tru, When we donot eject the null hypothe this verde doesnt mean atthe ‘elendortsinsecet tut song enough 10 e sre of ut Notreecng Hy doesnot mean nat tt means tat tbe evidence want sreng crouch reject * If theres suficiem evidence to reject the nll hypothesis, we conclude thatthe alternate hypothesis is tue. * Ifthere is not sufcint evidence to reject the null hypothesis, we conclude that the ull hypothesis might be trv, but we never conclude that the null hypothesis i true. _—_—_ Example 9.5 State a conclusion when the null hypothesis is not rejected Boxes ofa certain kindof cereal are labeled as containing 20 ounces. An inspector thinks that the mean weight may be less than this, so he performs a test of Ho: = 20 versus ‘Hy: w-< 20. He does not reject the null hypothesis. State an appropriate conclusion, nxt means that the evidence wasn't song enough to eect it. An appropriate way to state conclusion when the null hypothesis is not rejected isto state that there is ot cuficient evidence o conclude that His tru. Solution ‘The ull hypothesis isnot rejected, so we do nothave sufficient evidence to conelude thatthe alternate hypothesis is tue. We can express this as follows: “There is not enowgh evidence ‘oconclude thatthe mean weight of cereal boxes is less than 20 ounces.” Another way 10 state this is: “The mean weight of cereal boxes may be equal 1 20 ounces.”Objective 4 oistinuish ‘uetween Type | and Type EXPLAIN IT AGAIN Type | and Type fl Errors Whenever a decision is made, there is a possibility that it is the wrong decision. There are tywo ways to make a wrong decision with a hypothesis test. First, if Mp is true, we might mis- might sistakenly decide not to reject it. These two types of ertors have names. ejecting Hy when itis true is called a Type Kerror. Fail ing to reject Hy when iis falsc is called Type IT error. We summarize the possibilities he flowing tbe, takenly reject it, Second, if H is fase Reality HM, Tone Hy False Reject Tyyelenr Correct decision Doa't reject fig Correct decision Typ In general, a Type 1 ector is more serious than a Type If error, eeause a Type I error onclesio, while Type Il eroe results only in no conclusion being made. For exan a Type fervor occurs when an innocent defendant is found guilty, anda Type If ector oscars when guilty defendant is found not guilty. Ideally, we would like to minimize the probability of both errors. Unfortunately the only way todo this is to increase the sample size, which in practice is often not possible With a fixed sample size, decreasing the probability of a Type T error increases the prob- ability ofa Type Il error, and vice versa, To see this, note that if we want to reduce the probability ofa Type I errr, we must inetease the strength of the evidence needed to reject the null hypothesis, making it less likely that we reject Hp when itis true, However, it will then also be les likely that we reject Hg when ti false, which will increase the probability ‘of a Type Il error. Similarly if we reduce the strength of evidence needed to reject Ho, We decease the probability ofa Type I error but increase the probability of a Type I error. Because a Type I eroris usually mor serious than aType I error, hypothesis tests are often designed so that the probability of Type I eror willbe aceptably small, often 0.05, 0 0.01. This values called the significance level of the test, which we will discuss further in Section 92. results in a fal Se, in a trial Example 9.6 Determining which type of error has been made The dean ofa business schoo! wants to determine whether the mean starting salary of grad- uates of her school is greater than $50,000. She will perform a hypothesis test with the following noll and alterate hypotheses: Hg: w= $50,000 Hy: p> $50,000 8, Suppose thatthe true mean is 1 = $50,000, and the dean rejects Hg. Is this a Type I error, a Type I error, ora correct decision? b. Suppose thatthe true mean is jv = $55,000, and the dean rejects Ho. Is this a Type 1 error, a Type II error, ora correct decision? ¢. Suppose thatthe true mean is 4 = $55,000, and the dean does not reject Ho. Is this Type I error, a Type Il error, ora correct decision?