Please enable JavaScript.

Coggle requires JavaScript to display documents.

Unit 4: Producing Data Concept Map (Observational Studies and Experiments,…

- - - - Choosing the easiest possible individuals, in order to reach results quickly.
      - If I wanted to create my own survey to see who MA favored to become their next Senator when one of the current ones resigns/moves to a different position, my just sampling people from Duxbury would be bad. I would be working with a far wealthier and more educated sample than the general populace of MA.
    - - A type of sample consisting of people who have chosen themselves by responding to a general appeal, often are very bias due to the fact that the people who respond usually have very strong opinions.
      - The shopping channel wants to know how often people actually buy things advertised on their channel, so they post a poll on their website asking whether the respondent had purchased an item. This is not representative of the percent of Americans who have bought something from a shopping channel, since the people visiting the channel's website are far more likely to have bought something.
    - - Undercoverage is an affect of certain groups being left out of the sample choosing process.
      - If I wanted to do a phone survey of Massachusetts residents on whether they like the current Governor Charlie Baker or not, but my system only allowed calls to landlines and not cell phones, I would be leaving out a significant population: anyone who owns a cellphone but doesn't have a home phone would be left out.
  - - - Sometimes a survey question may intentionally or unintentionally
        word questions in such a way that it implies an answer. A similar problem in journalism is called a "leading question," which implies an answer.
      - For example, consider this question from a University of Denver survey about the Holocaust: "Does it seem possible or does it seem impossible to you that the Nazi extermination of the Jews never happened?" This question is intentionally confusing and misleading. 22% of people answered in such a way that implied they believed the Holocaust was a fake. Conversely, a still disturbing, but much smaller, 9% answered in the affirmative to this: "Do you doubt that the Holocaust actually happened or not?
    - - A non-response error is one which the people who respond to the survey are not the exact same as those who are chosen, because they are unwilling or unable to do so.
      - Pew Research Center is one of the nation’s most trusted polling and survey companies, and they tend to produce accurate, non-partisan and pretty unbiased results. Even so, their non-response rate is 73%, which dilutes the originally chosen random sample and could skew results.
    - - Within a survey, people often will answer incorrectly, especially
        in an anonymous survey. People tend to give answers that they tend to feel are more socially acceptable or fitting rather than actual opinions.
      - Some have suggested that response bias is an
        explanation for why Donald Trump was able to poll at such low levels prior to the election but still win it. People may have not stated their preference for him because it was socially awkward to support such a flippant candidate, but still ended up voting for him on election day.Attached is a screenshot of the New York Times' errant election prediction.
  - - - One advantage of this is that it creates results which are representative of the whole population. Whereas in a SRS you may get draw an imbalanced proportion from each strata, in this results can be made to be perfectly proportional to the population's makeup.
      - A type of sample in which groups, also called strata, of similar individuals, are created from the entire population so that an SRS can be performed on each group. These results are then combined to create the full sample.
      - Much like the SRS, the Stratified Sample is neither cost-effective nor timely. The strata must be separated and then a random sample from each must be drawn, taking up much time and money.
    - - Similar to a Stratified Random Sample, a Cluster Sample involves division by groups of the population. A sample is taken from each of these so-called "clusters," which are a group within the population which are geographically close within the sample.
      - Cluster sampling’s main advantages lie in its’ efficiency and cost-effectiveness. In situations where it would be prohibitively costly to, or time did not permit, travel between swathes of the general population, cluster sampling is quick and helps give a decently accurate picture of the group.
      - Cluster sampling’s disadvantages lie in its’ imprecision. Clusters aren’t really going to be perfectly representative of the general population, so in this technique a little precision is compromised for convenience.
    - - By using SRS you are creating an equal opportunity for every sample to be chosen, which is an obvious advantage. In basically every other method discussed, there are some individuals which may have a higher chance to be chosen, in this it is truly random.
      - This is a very time consuming and cost-inefficient sample, especially in situations where time or money might be tight.
      - A sampling technique in which a certain amount of individuals out of the entire population are randomly chosen, this is done in such a way that every individual has an equal chance of being chosen by assigning each individual a set of identifying digits and choosing random numbers.
      - I want to survey Duxbury juniors about whether they have taken the SAT yet and how many times they plan to take it. I will assign each member of the Class of 2019 a random identifier based on the unique last 3 digits on their Student ID, and use the random integer function on my calculator to pick out 30 random people.
  - - - The student council wants to gauge interest for a spring dance, but rather than go through the trouble of pushing a poll to all 1,000 students, they'll use a sample of 105 students. Knowing that there may be a difference in opinions between grades, they decide to use a cluster sample with SRS.
        
        Since every student is required to take a gym class, and gym classes generally involve a representative mixture of grades and genders, the clusters will be taken from gym class. From each period, with 150+ people in the gym, a random sample will be taken from these clusters.
        
        Using the teachers’ attendance sheets, the representative will assign each of the kids a random number between 1 and the 150 students in each period based on alphabetical order. They will use the random integer function on their calculator, and pull the first 15 digits, excluding any repeat results, and poll those students. This process will be repeated for all 7 periods.
        
        The use of the SRS here is to cut down the sample to a group which can be interviewed to show the preferences of the school without interviewing everyone.
- - - - For example, measuring how the factor of height growth affects a student’s reading score would have to be an observational study, since a researcher clearly cannot control a child's growth or randomly assign "growth"
      - Another example: measuring how a mother's intake of calories while a pregnant affects the baby’s birth weight. A researcher can hardly force a mother to consume a certain number of calories, for ethical reasons, so this would be a study, not an experiment.
    - - For example, a drug company wants to test the effects of their new concoction on their rats. They random assign a group of 50 rats, half to their old medicine, and half to their new. This would be an experiment, since the researcher is able to randomize factors.
  - - - The researcher must control for such things as lurking variables, by setting up a group which receives no, or completely neutral, treatment. For example, if a researcher was testing the effects of a cold drug, they might give half the group an older drug to see if there was any improvement, or might control for gender by giving half of the males the new drug, half the old and the same for females.
    - - The experiment must have enough experimental units to overcome the natural variance in the outcome. For example, if a drug company is testing their drug and only uses two rats, it is possible that one of them might have a natural resistance to the drug, which would seriously skew the results. Performing the same experiment with dozens or hundreds or rats would correct for this.
    - - The factors, which are the different treatments (or explanatory variables) assigned to the experimental units must be done so randomly. If the assignment is not done randomly it cannot be called an experiment, only an observational study.
  - - - Blinding and double blinding is the practice of not telling the participants they are receiving a control, and not telling the experimenter if the subject is receiving control, respectively. This is important because it can help diminsh effects of placebos. Double blinding provides for even more of a failsafe, because it prevents the researcher from accidentally tipping off the fact that it is a placebo through body language or a slip of the tongue.
    - - If subjects respond differently after given a treatment they are told is real, even though it is neutral, this is the placebo effect. The majority of placebos occur in drug trials, where people may report improvement in their conditions even when they are given the control sugar pill.Placebos must be controlled for.
        
        Possibly the most hilarious placebo of all time, even though it wasn't in an experiment.
        Residents in South Africa got rashes and other sicknesses right after a new cell tower was installed, so blamed it on the tower. Hearing these accusations, the provider switched off the tower without telling anyone. At the next court hearing, the residents complained of symptoms getting worse, so the company revealed that the tower had been off since the first complaint. Most people quickly stopped attending the hearings thereafter.
  - - - Our subjects- 132 people who reported having significant cold symptoms of various ages.
        
        We will divide them into six blocks
        
        22 Female Children (0-18)
        
        11 Receive Sugar Pill
        
        11 Receive 20mg of new drug
        
        22 Male Adults (18-65)
        
        11 Receive Sugar Pill
        
        11 Receive 20mg of new drug
        
        22 Male Children (0-18)
        
        11 Receive 20mg of new drug
        
        11 Receive Sugar Pill
        
        22 Female Adults (18-65)
        
        11 Receive Sugar Pill
        
        11 Receive 20mg of new drug
        
        22 Male Senior Citizens (65+)
        
        11 Receive Sugar Pill
        
        11 Receive 20mg of new drug
        
        22 Female Senior Citizens (65+)
        
        11 Receive Sugar Pill
        
        11 Receive 20mg of new drug
        
        Among each of these groups we will use a SRS to divide them in half
        
        This experiment will be double blinded, so neither the subjects nor the nurses know which treatment each subject receives
        
        Placebo is account for through the use of the sugar pill, which would control for those who might seem to feel better through mental impacts, not the drug's affect.
        
        Experimental Design Vocabulary
        
        1 more item...
        
        Response Variable: Response to question “On a scale from 1-10, how did your symptoms abet (10=completely better 1=no change)