Exam 1, Biology 301C, 14 September 1998

Language of evaluation

1,2. (3.5 pts each) We wish to test the model that red cars have at least 1.5 times higher accident rates (per car per mile) than white cars. For the following possible data, which consequences (A-D) apply? Note that this question is not about causation versus correlation, only about data and a model that happens to describe a correlation. Answer each question independently of the others. At least one answer, but possibly more (MTF). Ignore the possibility of sampling error.

A) The data are inconsistent with the model

B) The data are consistent with the model

C) The data support the model

D) The data are irrelevant to the model

E) None

1) Data: Accident rates by color of car per 10,000 miles:

white = 1%, blue = 0.6% tan = 0.7% green = 0.4%, red = 1.5%. A B C D E

since rate of red = 1.5X that of white, is both consistent and supports (since they could have gone the other way)

2) Data: Accident rates by color of car per 10,000 miles:

white = 1%, blue = 0.6% tan = 0.7% green = 0.4%, red = 2%. A B C D E

same logic as in (1): red = 2X that of white, which is also consistent with and supports the model

Correlations & Causation

3. (8 pts) Which of the following statements describe a (non-zero) correlation? Do not choose any option that describes a zero correlation or for which a correlation is undefined. If insufficient information is given to determine whether a correlation exists, treat it as if there is no correlation. MTF

(A) More people attend the Kerrville Folk Music Festival than attend the Rice Festival.

two variables (festival and # attending) and the number attending differs; so = a correlation

(B) Two thirds of UT students voted in student government elections; one third did not vote.

only one variable here, number of students voting (gives same info as # not voting). Need another election for a second variable.

two variables: temperature and time; not constant, so = correlation

(D) More people buy products endorsed by Michael Jordan than products endorsed by O.J. Simpson two variables: celebrity and number buying products; number buying products differs between the celebrities, so = a correlation

(E) The world’s population has doubled in the last 60 years. two variables: time and population. The two change together, so = a correlation

(F) Customer sales at Austin Restaurant Supply have been flat (unchanged) over the last 7 years. two variables: sales and time. However, sales have not been changing, so a zero correlation. You could also argue that there is only one variable, if you consider that, since sales have not changed, that is not a variable.

(G) Retail sales in the U.S. are higher in December than in any other month. two variables: month and sales. They change together, so = a correlation

(H) Death rates for skydiving are 1 in 100,000 jumps; death rates for driving are about 1 in 170,000 miles driven. one variable: death rate. the other data do not constitute a variable, because jumps differ from miles driven. So you don’t have two variables here.

4. (5pts) Key code, name, and ID number. Fill in (A B) in scantron field 4 to indicate your key for this version of the exam. Most everyone gets this correct, but failing to bubble in everything correctly costs points.

Be sure your name and personal ID number are correctly bubbled in on the scantron.

Your name is required on this exam form and the scantron form to receive credit for this test.

5. (7 pts) Wording has changed!!! Mark all models(s) that are inconsistent with the information in the following graph. That is, mark an answer if it CAN be ruled out using the information in the following graph. Assume you have no data other than what is presented in this graph. MTF

(A) Student GPA is correlated with hours of activity outside of class

(B) Student GPA is negatively correlated with hours of activity outside of class

(D) Getting a high GPA motivates a student to give up activities outside of class

(E) Getting a high GPA motivates a student to take up activities outside of class

(F) Taking up activities outside of class competes with study time and lowers a student’s GPA

(G) Taking up activities outside of class increases a student’s GPA

(H) Fluoride causes changes in tooth decay.

the graph shows a simple negative correlation. Options D-H are causal models, and because of the third variable problem, any causal model is considered to be consistent with simple correlations. So they would not be circled (you should only circle ones that the graph can rule out). Options A, B, C are statements of correlation, and only C goes against the graph. So only C is inconsistent.

6) (4pts) Consider a correlation between variable Y and diabetes. If X (not Y) is the factor that affects diabetes rate, which diabetes rates are expected in cells 1 & 2 of the following table? Assume that no other variables besides X and Y are important. (one answer only)

		Y:
		absent	present
X:	absent	(1)	low diabetes
X:	present	high diabetes	(2)

A) 1 is high, 2 is high	C) 1 is low, 2 is high
B) 1 is high, 2 is low	D) 1 is low, 2 is low

If X is affecting the data, then the data will be the same if X is the same. Thus (1) will be the same as the other cell in its row (low) and (2) will be the same as the cell in its row (high). Answer is C.

7. (6 points) Which of the following constitutes an example of inferring causation from correlation (i.e., in which a correlation leads someone to infer the causal basis of the correlation)? Base your answer only on the information provided. Do not circle answers that merely describe a correlation, that infer correlation from causation, or that test the causal basis of a correlation. MTF

(A) The Centers for Disease Control observes that countries with a high content of fish in the diet have lower heart disease rates than countries with low fish content in the diet. They then divide 1,000 U.S. citizen volunteers into one group that is fed 1 pound of fish/week and another group that is fed only 1 pound of fish/month. First sentence describes a correlation. Second describes an experiment to evaluate the cause. Not an answer to the question.

(B) Quitting the smoking habit reduces a person’s lung cancer rate. As a consequence, former smokers who have quit the habit have lower lung cancer rates than those who continue smoking. First sentence is causation. Second is correlation. Not an answer.

(C) People who drink modest amounts of alcohol have higher survival rates than people who drink nothing and than people who drink excessively. As a consequence, the medical profession is now beginning to suggest that modest alcohol consumption is a way to enhance longevity. First sentence is correlation. Second is a recommendation based on presumed causation. This one should be bubbled in.

(D) The global average temperature has increased over the last 50 or so years, following increased output of so-called greenhouse gases (carbon dioxide, methane) produced by human activities. Anti-environmentalists argue that the global warming is independent of and has nothing to do with human activities. First sentence is a correlation. Second describes people who argue that there is no casual basis to the correlation. Not an answer.

8. (6 pts) Consider the correlation that:

people in cities and towns with high fluoride in the water have low tooth decay rates

Which of the following causal models of this correlation use a “third variable” to explain the correlation? Use the method given in class to distinguish whether a 3^rd variable is invoked. Do not mark an option if the correlation goes in the wrong direction (e.g., that cities of high fluoride have higher levels of decay). MTF

Causal model	Cause invoke a third variable?
water with high fluoride also has high magnesium, and high magnesium reduces tooth decay Assumes a third variable, magnesium. Changing fluoride but not changing magnesium will not affect tooth decay.	Fill in (A) if a 3^rd variable is invoked fill in (A)
high fluoride inhibits the growth of bacteria that cause tooth decay, and it is the lower levels of bacteria in water with high fluoride that result in lower tooth decay rates. Fluoride causes changes in bacterial levels that cause tooth decay. Changing fluoride will change tooth decay. No third variable.	Fill in (B) if a 3^rd variable is invoked leave blank
water with high fluoride comes from sources that also have fewer bacteria that cause tooth decay, and it is the lower levels of bacteria that result in lower tooth decay rates. Assumes a third variable, bacteria. Fluoride is merely correlated with bacteria, and changing fluoride levels in this model will not affect tooth decay because the bacterial levels are not affected (caused) by fluoride levels.	Fill in (C) if a 3^rd variable is invoked fill in (C)
high fluoride makes teeth resistant to tooth decay by affecting the properties of enamel. Fluoride causes enamel changes that affect tooth decay. Changing fluoride will change enamel that will change tooth decay. No third variable.	Fill in (D) if a 3^rd variable is invoked do not fill in (D)

9-11. (3.5 points each) Gotham City has a long history of upholding high moral standards, and its residents have consistently voted for high beer taxes. Mayberry residents like beer and have consistently voted to keep beer taxes low. Researchers have discovered that Gotham City has high STD rates in teenagers, whereas Mayberry has low teen STD rates. Residents of Gotham City have thus proposed lowering beer taxes as a way of reducing STD rates.

Use the following variables:

Variable 1: STD rate

Variable 2: beer tax rate

Variable 3: city

For each of the following questions, you are given a pair of these variables. You are asked to choose among the following 3 options that best characterizes their relationship in the problem description above.

(A) no correlation or causation is indicated.

(B) a correlation is indicated, but no causation between the variables is suggested

For each pair of variables given below, which option applies (one answer each)?

9. Variables 1 & 2: (A) (B) (C) Correlation described, but problem goes on to say that residents have voted to increase taxes to lower STD rates. Thus a causal relation is suggested.

10. Variables 1 & 3: (A) (B) (C) Correlation described only.

11.Variables 2 & 3: (A) (B) (C) Certainly a correlation. But problem also says that residents have voted for high or low taxes. So a causal relationship is described as well.

Controls

12. (5 pts) The Monty Python video on penguin intelligence compared the performance of humans and penguins on an IQ exam. Consider the last test shown in that video (with the immigrants at the zoo). Mark all of the following factors that were controlled for in that IQ test shown (recall that a factor controlled for is one that is matched across the different groups being compared).

(A) inability to speak English	(D) body size
(B) ability to speak English	(E) testing environment
(C) brain size	(F) environment in which the subjects were born and raised

The last test was of penguins versus non-English speaking people. Thus both ability and inability to speak English was controlled. Brain and body size was not controlled, because both organisms were their natural size. Testing environment was the same (zoo), but they were not both raised in the same environment.

13,14. As described in the Book, epidemiologists in Britain noted a correlation that certain cancers were more frequent among residents living near nuclear power plants than in the population at large. The following two questions pertain to this study and its implications.

13. (4pts) Which of the following models are consistent with this correlation? MTF

(A) nuclear power plant locations reduce cancer rates, but the people who live in these locations have ethnic cultures that elevate their cancer rates

(B) nuclear power plant locations have no effect on cancer rates, but the people who live in these locations have ethnic cultures that elevate their cancer rates

all are consistent because they are all causal models, and any causal model is consistent with a simple correlation. The reason is that hidden variables may be present.

14. (4pts). Now suppose that we had been randomly assigning where people live in Britain over the last 100 years, and that we still observed that residents living near nuclear power plants had higher-than-average cancer rates. (Randomly assigning where a person lives would of course be unethical. However, assume for the sake of this question that it could be done.) Which of the following models would now be consistent with this correlation? MTF

(A) nuclear power plant locations reduce cancer rates, but the people who live in these locations have ethnic cultures that elevate their cancer rates

(B) nuclear power plant locations have no effect on cancer rates, but the people who live in these locations have ethnic cultures that elevate their cancer rates

In this case, we have a randomized experiment that will have controlled for hidden variables. Now the only model consistent with the data is (C), that the locations cause the higher cancer rates.

15. (5 pts) Each of rows (A)-(G) describe different treatments that could be applied to cornfields in attempting to maximize yield. The treatments differ in which factors are present (indicated by “+”) or absent (-). Factor 1 is fertilizer; factor 2 is insecticide; factor 3 is herbicide; factor 4 is plowing before planting; factor 5 is use of genetically modified seeds. For each treatment, the data gathered are the pounds of corn harvested per acre.

Which two treatments would you want to compare to determine if factor 1 is correlated with differences in corn harvested when all other factors are controlled? In evaluating possible answers, pick any comparison that controls for all unwanted factors, and assume that these treatments differ only in the ways stated. Mark exactly two options, or option I if none apply. Each row (each option) describes a different set of conditions, so to know which factors would be applied in a treatment, you look across the row. If multiple combinations satisfy the problem, any correct combination will be accepted. (Two answers or None; options have changed).

	factor
Option		1	2	3	4	5
	(A)	+	-	-	+	+
	(B)	-	+	+	-	-
	(C)	-	-	+	-	+
	(D)	-	-	-	+	-
	(E)	+	+	-	-	-
	(F)	+	-	+	-	-
	(G)	+	+	-	-	+
	(H)	+	+	+	-	-
	(I)	No combination satisfies the request

You want two rows in which (i) factor 1 is + in one and – in the other, and (ii) factors 2-5 are matched between the rows. Factors 2-5 within a row can be anything, but 2 must be the same between the rows, 3 must be the same between the rows, etc. (B) and (H) satisfy these criteria.

16. (5 pts) Control groups versus controlled factors. A professor conducts an experiment to determine how students can improve their exam scores. She uses two Sections of the same course that she teaches (each with different students) and lets the students in one section go about their business as usual (= the control group). For the other section (the treatment group), she dictates the students’ sleeping, eating, and studying habits for a week. She then gives the same exam to both sections at their usual times and compares the scores between 50 randomly chosen male students of one section to the scores of 50 randomly chosen males from the other section. What factors are explicitly controlled for (matched) in the design of this study? Do not infer more than is given. MTF

(A) Section	(D) student sleeping habit
(B) student gender	(E) time and day of exam administration
(C) student eating habit	(F) prior exam performance of the student

The only things explicitly controlled will be known to be the same between the two groups: gender

Experiments

17. (5pts) Which options about the in-class personality survey are true? MTF

(A) It was an experiment because the survey used a blind design. Not what makes something an experiment. No

(B) Most of the Bio301 students scored the personality description of themselves as reasonably accurate (a score of 3 or better in a range of –5 to +5). Yes, over 75% gave it 3 or better (last year and in 2008)

(C) A video shown in class performed an experiment with a mock horoscope that had several features in common with the Bio301 experiment. Yes, indeed the 301D experiment was inspired by the video.

(D) The personality descriptions in these experiments (our class and the video) were specific in many details about the person. The descriptions may have seemed specific but they were vague and general.

(E) The study shown in the video (and used in class) would have been improved by including a group in which the personality description was assigned randomly to the student. No. Everyone got the same personality description, so randomizing who got them would make no difference.

18. (6pts) Prisoners of Silence video. Which of the following options about the FC (facilitated communication) video are true? MTF

(A) Experiment: Lecture and the book described two types of experiments regarding how to control for unwanted correlations among variables. The FC experiment was the type in which unknown hidden variables were controlled for by randomization. No. Randomizing treatment versus control is done when the variables are unknown. Here the relevant variable was known, so randomization to get rid of unknown variables was not needed.

(B) Replication: tests were conducted with multiple autistic children, multiple facilitators, and the type of test was even varied. Yes, there were 3 or 4 levels of replication, including these.

(C) Controls: in these studies, the controls were the parts of the tests in which the facilitator and child were shown the same information. Yes, being shown the same information controls for possible effects of the testing environment. And they were sometimes shown the same info.

(D) Blind was an essential feature of these tests. Blind is the essence of the experiment – keeping the facilitator and child from seeing what the other saw.

19. (7 pts). Which of the following studies describe experiments, regardless of whether the experiment was designed well or poorly and regardless of ethics. In each problem, the goal is given. The question is whether the option describes an experiment with respect to the goal. MTF

(A) To test whether smoking causes lung cancer, you interview people about their smoking habits. You then take the smokers and pay a randomly-chosen half of them to quit, which they do. After a year (with demonstrated compliance to their group), you then look for an association between lung cancer and level of smoking. Yes, an experiment because you get some people to change their smoking habits to see if lung cancer rates change.

(B) To see how observant your friends are, you part your hair on the left side one month and then switch the part to the right side the next month. You keep track of who notices the change. Yes, an experiment. Again, you have manipulated something out of the ordinary to see what happens.

(C) To figure out how to make the ideal elk steak for your guests, you compare the recipes of different chefs whose steaks you have tasted in the past. You choose the recipe that is simplest but whose steak also met your standards of excellence. Not an experiment. You haven’t changed anything, merely looked at alternatives.

(D) Your mechanic replaces the spark plugs and spark plug wires in your car, and your mileage improves. You then change the spark plugs in your car back to the old set to see if the improved mileage is due to the new plug wires. A manipulation to evaluate a model (the cause of improved fuel efficiency). = an experiment

(E) You are in charge of a small airport in which two skydiving deaths have occurred in the past 3 years. You institute a mandatory checklist for all skydivers to complete before takeoff to see if the accident rate declines. An experiment – instituting the checklist to see if survival rates improve.

(F) In an effort to identify the causes of university student academic success, a researcher monitors the habits of university students and matches those habits to GPA. Just correlational data. No experiment.

Random not on 2008 exam

20. (6 pts) You do a statistical test of the difference between the average values observed in a treatment group and a control group. Which of the following options about P-values are true? MTF

A) The P-value is the probability that the difference between the averages (of the treatment and control group) is due to something other than randomness (or, in other words, due to something other than sampling error). No, just the opposite. The P-value is the chance of getting the difference (or a more extreme difference) when ONLY sampling error is present.

B) A result of P< 0.95 is considered the minimal threshold for accepting the model that treatment and control group averages are not due to just randomness (sampling error). No. P< 0.05 is the threshold. If you accepted P<0.95, you would almost always be rejecting the null model.

C) A P-value less than zero indicates that the null model is true. Nothing – no value – indicates that a model is true.

D) Two studies with the same difference in averages (between the treatment and control group) will have the same P-value. No. The P-value also depends on the variance in the data, not just the averages.

E) P < 0.01 means that the null model of randomness would be expected to account for the observed results less than 1% of the time. Yes, the P-value is indeed how often you expect to get the results under the null model of randomness.

Causal model

Cause invoke a third variable?