Stat 512 by zhouwenjuan

VIEWS: 9 PAGES: 22

									Stat 512


    Day 2: Designing Experiments
Leftovers from Tuesday

   Questions on syllabus?
   Repeating the question
   Blackboard problems?
       Sooner turn in, sooner get feedback?
                                                      Figure 1: Comparison of Airway Obstruction
                                                                 from Level of Exposure


Last Time                                            100%

                                                     80%




                                        Percentage
                                                     60%                     43       airw ay not obstructed
                                                               52
                                                     40%                              airw ay obstructed

                                                     20%
                                                                             15
                                                                6
   Microwave popcorn factory                         0%
                                                               low          high


     Two categorical variables
                                                               Level of Exposure




           level of exposure, whether airway obstruction
       Graphical summary: segmented bar graph
       Numerical summary: difference in conditional
        proportions
       Not able to draw cause-and-effect conclusions
           Other differences between the two groups that
            might explain the higher airway obstruction rates in
            high exposure group?
       Not able to generalize to all microwave plant
        workers
           “Healthy worker effect”
                                                        100%




Last Time
                                                        80%




                                           Percentage
                                                        60%
                                                                                                           Control
                                                                                                           Lung cancer patient
                                                        40%


                                                        20%


                                                         0%
                                                               none   light     mod   heavy excess chain



    Smoking and lung cancer
                                                                              Level of Smoking


       Two categorical variables (amount of smoking and
        with or without disease)
       Graphical summary: segmented bar graph
       Numerical summary: difference in conditional
        proportions
       Can’t draw cause and effect conclusion
           Could be some other difference (diet) between the EV
            groups that explains higher lung cancer rates with more
            smoking
Last Time

   Generalizing from sample to population?
       Reasonable to conclude 605/(605+780), or 44%,
        of all males of similar ages and economic status
        have lung cancer?
   Measurement issues
       Relying on recall…
       Know they are sick…
Second famous smoking study

   Hammond and Horn (1958)
   Find 12,000 healthy men, complete a
    questionnaire on smoking habits, had 22,000
    American Cancer Society volunteers follow
    them for 44 months to see whether they die
    from lung cancer
       Advantages? Disadvantages?
Practice Problems

   Identifying variables
       Supreme Court Justices:
           Qualitative: gender, party
           Quantitative: age, number of yes votes
       Not OK:
           number of republicans
Practice Problems

   Victims of violence
       OU: people (not “number of victims”)
       EV: whether abused
       RV: whether commit crime
       Not ok in defining variables:
           Those who…
           Number abused…
           Whether abuse leads to violent crime…
       (c) Mostly to spur discussion…
           Which variable was “controlled” by the researchers…
More practice:

   The book Day Hikes in San Luis Obispo County by
    Robert Stone gives information on 72 different hikes
    that one can take in the county. For each of the 72
    hikes, Stone reports the distance of the hike (in
    miles), the anticipated hiking time (in minutes), the
    elevation gain (in feet), and the region of the county
    in which the hike can be found (North County, South
    County, Morro Bay, and so on, for a total of eight
    regions).
       Observational units?
       Types of variables
More Practice: Hiking in SLO

   Are these legitimate variables?
       Longest hike in the book
       Those hikes in the North County region
       Proportion of hikes with an elevation gain of more
        than 500 feet
       Is hiking time related to elevation gain?
Example 1: Near-sightedness and Night
Lights
   “Myopia and ambient lighting at night,”
       Quinn, G.E., Shin, C.H., Maguire, M.G. and
        Stone, R.A.
       Nature, 399:113-114, 1999.
Example 1: Myopia and Night Lights

                Room light                Night light          Darkness
Far-sighted     12                        39                   40
Normal          22                        115                  114
Near-sighted    41                        78                   18

               100%
               90%
               80%
               70%
               60%                                          Near-sighted
               50%                                          Normal
               40%                                          Far-sighted
               30%
               20%
               10%
                0%
                      Room light   Night light   Darkness
Example 1: Myopia and Night Lights

   Base rate: .286
   Conditional Proportions
       Room light: .55 myopia, .16 hyperopia
       Night light: .336 myopia, .168 hyperopia
       Darkness: .105 myopia, .232 hyperopia
   Convincing evidence that using more light in
    the child’s room causes a higher rate of
    myopia?
Terminology: Confounding variable

   Has an influence on the response, but its
    effects cannot be separated from those of the
    explanatory variable
             room light
             parents with bad eyes       Compare
                                        eye-sight
            darkness
            parents with good eyes
Example 2: Have a Nice Trip

   Can instruction in a recovery strategy
    improve an older person’s ability to recover
    from a loss of balance?
       12 subjects have agreed to participate in the study
       Assign 6 people to use the lowering strategy and
        6 people to use the elevating strategy
       Similar amounts of men in both groups?
           Proportion of group 1 that are male – proportion of group
            2 that are male
           What would we like to be true about these proportions?
Investigation 1-7: Have a Nice Trip

   Take 12 index cards
   Put one of the 12 names on each card
   Shuffle the cards and deal out 6 to learn the
    lowering strategy and the other 6 to learn the
    elevating strategy
   What is the number of males in each group?
   The male proportion in each (out of 6)?
   What is the difference in these 2 proportions?
Pool Results

   Quantitative variable
   Different type of graph…
       What are the observational units and variable in
        this graph?
Long-term pattern?

   Use applet to repeat the process a large
    number of times
   Open IE, double click on “Dr. Beth Chance” >
    “Statistical Methods” > “Stat 512 Java
    Applets” > “Randomizing Subjects
   The applet mimics exactly what you did with
    the index cards.
Effect of randomization

   If you randomly assign subjects to the
    groups, what is generally true about the
    groups?
   If after imposing the treatment, you later
    observe a difference between the groups, to
    what can you attribute that difference?
Moral

   Randomization equalizes variables between
    groups
       Should not have potentially confounding variables
       If later (after treatments) observe a difference
        between groups, feel comfortable attributing that
        difference to the explanatory variable
   Remaining question
       Will always be some difference, by chance
       How big does this difference have to be?
Example: Friendly Observers

   “The trouble with friendly faces: Skilled
    performance with a supportive audience,”
       Butler, J. L., and Baumeister, R. F.
       Journal of Personality and Social Psychology, 75:
        1213-1230, (1998).
To do

   By Friday, noon
       Turn in HW 1 (4 problems)
   By Tuesday
       Preview the Friendly Observers example
        (complete questions (a)-(e))
       Submit PP 2 in Blackboard
       Pick and start reading one of the 3 articles

								
To top