Please enable JavaScript.

Coggle requires JavaScript to display documents.

Unit 4: Surveys, Samples and Experiments (Samples and Surveys (Samples…

- - - - Vocabulary
        
        Under-coverage
        
        Definition: certain people or groups of people within the population are not included in the data collection
        
        Ex: Someone collects data on the favorite sport to play from US teenagers (14-18 years old). The surveyor picks 500 random students from high schools across America. However, this excludes teenagers that are homeschooled, teenagers that have dropped out of high school, or teenagers that have graduated high school or gone to college early.
        
        Convenience Sampling
        
        Definition: using a group of individuals that is easiest to obtain
        
        Ex: Instead of choosing 100 random graduates from colleges across the entire world, the surveyor decides to survey 100 random graduates from Harvard University, a college 10 miles from them, to determine what the average starting salary after receiving a 4-Year Bachelor Degree
        
        Voluntary Response Sample
        
        Definition: a sample of data from a group of individuals who choose to respond to a survey
        
        Ex: After every phone call with Verizon Customer Service, an automated Verizon Customer Service Survey is sent out to the customer. However, the majority of the people who respond to these automated surveys either have a terrible Customer Service experience or a phenomenal one, so it creates an exclusive sample.
        
        Note: there is often bias in voluntary response samples, since there are many factors in why a person would rather respond or not. If a person chooses to not give their response, it is often because they do not have a strong opinion about the survey question. This would result in an inaccurate representation of the population.
    - - Stratified Simple Random Sample
        
        Definition: a sample of an equal representation, called strata, of the overall population. Take each of the groups in a population, and then take an SRS of each of the smaller groups of populations. (always identify the strata and which group is in the strata)
        
        Advantages
        
        Every group with different opinion is heard.
        
        It is much easier and faster to go group by group.
        
        Disadvantages
        
        The strata has a risk to not be proportional of the entire population.
        
        The responses within the strata may not be similar to the whole as only a sample of the group is taken.
      - Cluster Sampling
        
        Definition: the entire population is split into sections with characteristics with equal representation of the characteristics of the entire population
        
        Advantages
        
        Very easy to gather the data from the sample of the population.
        
        Disadvantages
        
        There will be groups of the population imperfectly accounted for
        
        It is not completely random (there is a certain cluster assigned)
      - Simple Random Sample (SRS)
        
        Definition: a sample in which any individual of the sample or any possible sample has the same chance to be selected
        
        Advantages
        
        It is the only sampling technique that is entirely random. Every individual has the same opportunity to be chosen as any other individual in the population
        
        It is entirely unbiased
        
        Disadvantages
        
        It is really difficult to actually perform sample random sample of big populations. It is also very time consuming
        
        You could unintentionally, get undercoverage within the data.
      - Click Here for the Properly Designed SRS
  - - - Response Bias
        
        Definition: certain individuals or groups of people in the sample feel compelled to answer the survey either falsely or deceitfully due to some sort of pressure
        
        Ex: The National Organization for Women, a prominent feminist group in America, stands outside of City Hall right before the elections in the Massachusetts primaries. The co-founder asks every single person who walks by whether they support equal pay between men and women across all professions. You, a man, who believes men are stronger and smarter than women, and therefore should be paid more in certain situations, say that you do support equal pay. You purposely answered with an untrue response so you wouldn’t receive backlash from the co-founder of this feminist organization.
      - Wording Bias
        
        Definition: phrasing a survey question with context or diction that influences the response of the individuals
        
        Ex: In this example, the two different wordings of the same question could change the answer:
        Since there are many poisonous spiders in America, causing hundreds of deaths a year, should the government develop a pesticide to kill all spiders in the country?
        Since spiders kill harmful insects like mosquitoes and wasps after trapping them in their web, should the government be allowed to create a pesticide to kill all spiders in the country?
      - Non-response Bias
        
        Definition: Individuals in the sample, for some reason, do not want to or cannot answer the survey
        
        Ex: A bald man with a Confederate Flag Shirt is standing outside of Home Depot, asking people whether they like dogs or cats better. Every single black man and woman walk right by the man, who clearly appears to be a white supremacist. The survey, therefore, was not an accurate representation of the population, since black people did not feel comfortable answering the question.
      - Bias
        
        Definition: a systematic error where a certain result of the survey has higher odds to be chosen than other ones, resulting in poor data
        
        Ex: The boss of your company needs to determine whether the amount of vacation granted by your company is adequate for the sales team. He stands outside of the cubicle floor (where every single sales employee was working that day), and asks every single person who exits the building whether they think the vacation time is enough. However, it soon is apparent there is response bias. Some people answer untruthfully that the vacation time is sufficient because they’re in a hurry to get home, and do not want the boss to feel compelled to ask further questions. Other people have been hoping for the sales manager promotion, and do not want to answer in a way that would harm their chances for the position. Because of this, the data is heavily skewed to the response “Yes” (there is enough vacation time), rather than “No” (there needs to be more vacation time). This would under predict the number of people who actually had an opinion that there should be more vacation time. Because of this, the boss may no longer feel incline to bring the issue of vacation time in the quarterly board meeting. Many people who may have wanted more vacation were not represented in the biased results; therefore the decisions made based on the results of this survey would not be well-informed.
        
        Note: The source of bias can come from a sampling or non-sampling error, and whenever there’s bias: identify it and identify the direction of the bias
        
        The direction of bias is the result in which the bias favors/leans to
- - - - A lurking variable is a factor in the experiment that is not a part of the explanatory or response variables, but it still has an effect on the outcomes of the experiment. A lurking variable can lead to confounding, a situation where the effects of two variables on the response variable can not be distinguished. If there’s a hanging (lurking) variable that is influencing the response variable, and there’s association between this lurking variable and the defined explanatory variable, confounding will occur.
        
        A project initiator for Panera Bread needs to create a new sales pitch for all of the 2019 advertisements to be based on. He decides to look into the relationship of number of Panera Breads to the number of food-borne illnesses that led to death. This project initiator decides to survey a proportionate number of countries from each continent based on population size. At each country, the number of Panera Breads as well as the death toll based on food-borne illnesses. After collecting and creating a model of the data, the initiator notices a strong negative correlation; as the number of Panera Breads increases, the number of food-borne illnesses decreases. He brings it back to Panera Bread headquarters, proud of his work. However, the lead statistician notices a problem immediately: there are lurking variables that have led to confounding. The issue, this statistician explains, is that the number of Panera Breads in a country is largely based on GDP and the access to healthy, sustainable food sources. Since these lurking variables are connected with the explanatory variable, number of Panera Breads, there is no way to indicate what is causing the relationship in food-borne illnesses in those countries.
    - - A situation where the subjects (who know that the experimental variable is being tested on them) respond differently due to their knowledge of treatment. Researchers have proven that individuals, even when given neutral treatment, still often respond differently, despite not being given any true variable.
        
        Your job is to taste test Stop N Shop (SnS) Ketchup compared to a popular brand name Heinz Ketchup. You set up samples of clearly-labelled Heinz Ketchup on the left side of a sample table, and clearly-labelled SnS Ketchup on the right side of the table. All day, people come up to taste both ketchups, and the results are heavily skewed towards Heinz Ketchup. However, you reveal at the end of the survey, that Heinz actually manufactured SnS Ketchup with the same ingredients and equipment, so the ketchups were identical. People responded favoring the Heinz ketchup due to the placebo effect, as they believed that the brand-name ketchup was better SnS ketchup based on the name only.
    - - Randomization
        
        Using random assignments, such as a random number table, to select the subjects for the experiment. Similar to above, it gets rid of any natural irregularities with the experimental units.
      - A Control
        
        The group in the experiment that does not receive the treatment. This creates something to compare the final response variables which creates a comparative experiment. This also controls the lurking variables that could be found in the experiment.
      - Replication
        
        The repeatability of the experiment. This is important because it means that the experiment has enough experimental unit to overcome any natural irregularities that could be found
    - - Experimental Factors
        
        The explanatory variables; the factors that the response explains the response variable
      - Response variables
        
        The variable that is explained and measured based on the experimental treatments on the subjects
      - Experimental Units (Subjects)
        
        The individuals that the experimenter assign the treatment to
    - - An attempt to overcome the placebo effect, by keeping information of the experiment from the individuals that the variable is being tested on and/or the observers documenting the effects of the variable. Keeping the information from the people being tested on only is known as single blinding, and keeping the information from both the people tested on and the observers is known as double blinding.
        
        You redo the Stop N Shop Ketchup taste test. This time, you do not clearly label the two different ketchup brands. Every time somebody comes over and tastes the two ketchups, they give their response first, without knowing the brand. Then, you pull off the covered label to record whether the brand was Stop N Shop or Heinz. This time, the results are nearly even among the two ketchups, indicating that the individuals responded differently when they knew the brand. Now, the results were more indicative of people’s true opinion of taste.