Utts Ch 5: Experiments and Observational Studies

1
Utts Ch 5 Experiments and Observational Studies

• Goal Learn to read news reports of studies
  critically.

2
Definitions (review)

• random experiment
• observational study
• explanatory variable
• response variable

3

• Do children who watch a lot of TV do poorly in
  school?
• Performance in school is the
• A explanatory variable
• B response variable

4

• Does a copper bar expand when heated?
• Temperature of the bar is the
• A explanatory variable
• B response variable

5

• Do people who drink more milk get sick less
  often?
• Frequency of illness is the
• A explanatory variable
• B response variable

6

• In general, do heavier cars get worse gas mileage
  than lighter cars?
• MPG is the
• A explanatory variable
• B response variable

7
Definitions (pp 82-84)

• Confounding variable affect on the response
  cannot be separated from the affect of the
  explanatory variable.
• Interacting variable (a third variable)
  possible impact due to combinations of the
  explanatory variables on the response variable.

8
Suppose an obs study finds that people who use
public transportation to get to work have a
better knowledge of current affairs than those
who drive to get to work, but the relationship is
weaker for well-educated people.

• 1. Method of getting to work is
• Response variable
• Interacting variable
• Confounding variable
• Explanatory variable

9
Suppose an obs study finds that people who use
public transportation to get to work have a
better knowledge of current affairs than those
who drive to get to work, but the relationship is
weaker for well-educated people.

• 2. Knowledge of current affairs is
• Response variable
• Interacting variable
• Confounding variable
• Explanatory variable

10
Suppose an obs study finds that people who use
public transportation to get to work have a
better knowledge of current affairs than those
who drive to get to work, but the relationship is
weaker for well-educated people.

• 3. Level of education is
• Response variable
• Interacting variable
• Confounding variable
• Explanatory variable

11
Suppose an obs study finds that people who use
public transportation to get to work have a
better knowledge of current affairs than those
who drive to get to work, but the relationship is
weaker for well-educated people.

• 4. Whether a participant reads the newspaper is
• Response variable
• Interacting variable
• Confounding variable
• Explanatory variable

12
Public transportation knowledge of current
affairs 13 p 102

• Method of getting to work
• D. explanatory variable
• Knowledge of current affairs
• A. response variable
• Level of education
• B. interacting variable
• Whether or not a participant reads the newspaper
  
• C. confounding variable

13
Basic principles experiments

• Randomization
• Placebos
• Control group
• Single-blind/ Double-blind
• Matched pairs, blocks, repeated measures

14
Random sample vs random assignment

• Random sample used to get a representative
  sample from the population of interest
• Random assignment used to control for
  confounding variables and other sources of bias

15
Difficulties and disasters in experiments

• Confounding variables
• Interacting variables
• Hawthorne, placebo and experimenter effects
• Ecological validity/ generalizability

16
Observational Studies

• prospective
• retrospective
• case-control studies

17
Difficulties and disasters observational studies

• confounding variables
• convenience or volunteer samples not
  representative.
• retrospective studies may require the past to be
  remembered.

18
Impotence Drugs Linked Again to Vision
ProblemsBy Alan Mozes?HealthDay 1/18/06
McGwin and his colleagues reviewed the health and
habits of 78 men who sought care at an
ophthalmology clinic in Alabama between 2000 and
2004 the authors found that patients taking
Viagra and/or Cialis who also had a history of
heart attack were almost 11 times more likely to
develop NAION than patients who had no similar
heart history and did not take either medication
The authors stress, however, that the risk for
these patients was weaker, only approaching
borderline statistical significance. And they
cautioned that the strength of the observed
associations -- if they hold up under future
studies -- remains, as yet, unclear.
19
Impotence Drugs Linked Again to Vision Problems

• What type of study is this?
• Experiment
• Observational
• Survey

20
Reflecting on values may counter stress by Amy
Norton, Reuters, 11/25/2005
Creswell and his colleagues had 80 college
students go through the stressful task of
explaining why they were a good candidate for a
university job to a pair of stern interviewers.
Beforehand, about half of the students answered
questions that caused them to think about an area
of life that they had earlier identified as
important to them -- such as religion, social
issues or politics. The rest of the students
answered questions on issues that were not
personally meaningful to them. Before and after
the stressful task, the researchers took saliva
samples from the students to measure levels of
the hormone cortisol, which tend to spike in
response to stress. Overall, the study found,
students who'd reflected on their personal values
showed less of a cortisol response to the
stressful situation.
21
Reflecting on values may counter stress

• What type of study is this?
• Experiment
• Observational
• Survey

22
Unfair boss could shorten your life study
Reuters, 10/24/2005
That crummy boss in the window office could be
slowly killing you The researchers tracked the
10-year incidence of heart disease in over 6,400
male civil servants in London who had been polled
on their perceived level of justice and injustice
in the workplace. "In men who perceived a high
level of justice, the risk of coronary heart
disease was 30 percent lower than among those who
perceived a low or an intermediate level of
justice," the researchers said. That finding was
not accounted for by other risk factors, from age
and socioeconomic status to cholesterol levels,
alcohol consumption and physical activity, the
authors said.
23
Unfair boss could shorten your life

• What type of study is this?
• Experiment
• Observational
• Case study

24
Study 7 Percent of Workers Drink on Job By
CAROLYN THOMPSON, Associated Press 1/11/06
Just over 7 percent of American workers drink
during the workday -- mostly at lunch -- and even
more, 9 percent, have nursed a hangover in the
workplace, according to a study. Young, single
men are tied most often to workplace-related
drinking, especially managers, salespeople,
restaurant workers and those in the media,
according to the findings by the University at
Buffalo's Research Institute on Addictions. The
results, culled from telephone interviews with
2,805 employed adults from January 2002 to June
2003, appear in the current issue of the Journal
of Studies on Alcohol.
25
7 Percent of Workers Drink on Job

• What type of study is this?
• Experiment
• Observational
• Sample survey

26
Drinking, Driving, the Supreme Court

• Craig v. Boren, 429 U.S. 190, 1976
• Challenged Oklahoma state law prohibiting sale of
  3.2 beer to males under 21 but allowed it to
  females under 21.
• Laws allowed to use gender-based differences if
  serve important governmental objectives and
  are substantially related to achievement of
  these objectives.
• Defense argued traffic safety was an important
  governmental objective and data show young males
  more likely to have alcohol-related accidents
  than females.
• Supreme Court shown two sets of data.

Source Gastwirth, 1988, pp. 524-528.
27
Drinking, Driving, the Supreme Court
2 of males 18-21
0.18 of females 18-21
28
Drinking, Driving, the Supreme Court

• 1. Determine if research was sample survey,
  experiment, observational study, a combination,
  or based on anecdotes.

Table 1 observational, representing only a
subset of those who committed the crimes. Table
2 sample survey, based on a convenience sample.
2. Are there any of the difficulties and
disasters inherent in that type of research?
Table 1 males more likely to be stopped for
other traffic violations. Table 2 drivers
questioned at these certain locations may not
represent all drivers
29
Drinking, Driving, the Supreme Court
3. Are there any alternative explanations for the
results?
Table 1 Males more likely to be stopped for
other traffic violations. Table 2 shows almost
80 of drivers were male. Males more likely
driving helps explain why Table 1 results so
different. Table 2 results of 11.4 males versus
9.4 females were statistically indistinguishable.
4. Determine if results meaningful.
Supreme Court overturned law concluding Table 2
data provides little support for a gender line
among teenagers.
30
Assuming the Supreme Court did take these data
sets into account to make their decision, why did
they determine the law to be unconstitutional
when each data set appears to favor those
defending the law?

• The first data set contains too many confounding
  variables to be helpful, in particular the fact
  that more men than women aged 18-21 were probably
  driving at that time, and were stopped more
  often.
• The second data set does not contain a random
  sample so it may not represent the population
  even if it did, the percentages are statistically
  indistinguishable.

31
Case summaryFour hundred women participated in
this study. Researchers found that children ages
3 and 4 whose mothers smoked 10 or more
cigarettes a day during pregnancy scored about 9
points lower on intelligence tests than the
children of nonsmokers. The gap narrowed to 4
points against children of nonsmokers when a wide
range of interrelated factors were controlled
for. The lead researcher was quoted as saying It
is comparable to the effects that moderate levels
of lead exposure have on childrens IQ scores.
32

• This was what type of study?
• Case-control design
• Observational study
• Matched-pairs experiment
• None of the above
• Which of the following describes difficulties
  with this study?
• Generalizability
• Confounding variables
• Possible experimenter effect
• All of the above

33

• What is one possible confounding variable that
  may or may not have been controlled for in this
  study?
• Possible answers socio-economic status exposure
  to secondhand smoke mothers diet, education,
  age, drug use parents IQ prenatal care
  duration of breastfeeding, etc.
• Based on this study, can a cause and effect
  relationship between smoking during pregnancy and
  childs IQ be established?
• Answer no. this was an observational study -
  people were not randomly assigned to either smoke
  or not many confounding variables could exist.
  But a cause and effect relationship cannot be
  ruled out either!