Chapter13 - Chapter 13 — Experiments &...

Info iconThis preview shows pages 1–4. Sign up to view the full content.

View Full Document Right Arrow Icon
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 2
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 4
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Chapter 13 — Experiments & Observational Studies * Skip: Adding More Factors (343-344) Definitions: Subjects (page 331) — the individuals being measured in the study. When the subjects are not human beings, they are usually referred to as experimental units. For example: rats are experimental units. Observational Study e A study that observes some characteristic about the subjects in the study, but does not impose any treatment. Observational studies are valuable for discovering trends and possible cauSe- and-effect relationships. For example, does smoking cause lung cancer? An observational study might measure the occurrences of lung cancer among a group of smokers versus a group of non—smokers and see if there is a difference in the rate of lung cancer. Note that the researcher is not forcing anyone to smoke. Instead, the researcher is just observing some characteristic among each group in the study. The researcher hopes to determine if there is a cause—and—effect relationship between smoking & lung cancer. While an observational study can help to identify a iink between two variables (such as smoking and lung cancer), a well-controlled experiment would be needed to prgyp a cause—and—effect relationship between the variables. Beware: Observational studies that are based on one’s historical recollection of events/facts often have errors. Observational studies are classified as either retrospective studies or prospective studies. Retrospective Studies — based on historical data The researcher collects data on past events only (pro-existing information). Because the data is historical, errors in memory sometimes occur. This, of course, can result in biased data. Prospective Studie — based on future events The researcher identifies the subjects in advance and then collects data as events unfoid. Experiment m A study that imposes some treatment on a group of subjects (or experimental units). For example, suppose a researcher wants to determine the effectiveness of two brands of blood pressure medication on reducing high blood pressure among middle-aged males. Cali the brands Brand A and Brand B. Furthermore, let’s say there are 800 middle-aged maie subjects in the study, all of whom have been diagnosed with high blood pressure. Haif of the subjects take Brand A medication daily for 1 year, and the other half take Brand B once a day for a year. At the end of the study, the researcher will compare the changes in the average blood pressure of the two study groups. Since the researcher is imposing some treatment, this is an experiment. Any properly designed experiment requires a random assignment of subjects to treatments. Furthermore, blinding is ofien used so the subjects do not know which treatment they are receiving. Blinding.(pages 339—340) Blinding is used to eliminate bias from an experiment. With blinded experiments, they can either be single—blind or double—blind. In a single—blind experiment, the person administering the treatment (medication, for example) is aware which subjects are receiving which treatment, but the subjects do not know. In a double-blind experiment, neither the person running the experiment nor the subjects know which treatment a given subject is receiving. The danger with making an experiment only single-blind is that the subjects might be able to discern which treatment they are receiving based on subtle clues that the experimenter is showing subconsciously. In our experiment, we are measuring blood pressure, which can be measured objectively with proper equipment. As such, blinding is not really necessary here. Imagine another study where the subjects were asked to rate (on a scale of 0 to 5) how they feel after taking a given medication. In this case, responses couid be highly affected if the subjects knew which medication they were receiving. Making this type of experiment double—blind would be highly suggested. 1 More Definitions (page 331) Explanatory Variable (the input) A variable that we manipulate so we can measure some related response. This variable is sometimes called the predictor variable. Often called the factor(s) of the experiment 0 Our experiment has one factor: Type of medication (Brand A or Brand B) Response Variable (the output) — a variable that responds to changes in the explanatory variables a In our experiment, the response variable is the change in blood pressure for each subject Levels of a Factor — The specific levels that the experimenter chooses for a factor 0 Our experiment has only one level — each subject is getting the same amount of medication. I Suppose 1/3 of those receiving Brand A medication were getting 5 mg/day, 1/3 were getting 10 mg/day, and 1/3 were getting 15 mg/day. Also suppose the same design is used for those getting Brand B medication. in this case, we still have only one factor (Brand A or Brand B medication), but the factor now has 3 levels (5 mg/day, 10 mg/day, 15 mg/day). Treatment — the combination of specific levels from all of the factors: (it of levels)n(# of factors) 0 For our revised experiment, there are 3 experimental treatments: (3 levels)(I factor) Experiments often employ the use of a control group(s). The purpose of the control group is to provide a baseline for comparison (page 339). We control sources of variation other than the factors we are testing by making conditions as similar as possible for all treatment groups. One commonly-used method is to use a placebo group in addition to the experimental group(s). Placebo (page 340-341) In the case of medicine, a placebo is just a fake medication that has no active ingredient. It looks/tastes/smells just like the real medicine, but it’s a fake. The use of a placebo is designed to control outside influences so the researcher can better identify the true relationship between the explanatory variable and the response variable. For our experiment, the researcher would be advised to also use a placebo group (a 0 mg/day pill). Thus, we would have another level to the given factor. (4 levels)(1fact°r} = 4 experimental treatments The benefit of using a placebo group here is because other stuff in each subject’s daily lives could have an influence on their blood pressure (not just the medication they are taking). At the end of the study, the changes in blood pressure for the experimental groups can be compared to the changes in blood pressure for the placebo group to see just how effective the medication is for lowering blood pressure. Statistically Significant (page 337) Are the differences witnessed from the experiment more extreme (bigger) than we would have expected just from random sampling variability? Because there are different subjects in the study groups, we must expect some differences between the groups just due to sampling variability. But, are the differences big enough to conclude- that taking Brand A or Brand B medication is more effective than taking no medication at all? If the differences are “big enough”, we say that the study results are statistically significant. This means that the differences between the groups can be attributed to the medication and notjust random sampling variability. Let’s assume a statisticaily significant difference is found between taking the medication versus getting the placebo. What is the scope of the conclusions that can be made? In other words, what is the population that we can draw conclusions about? When assigning which subjects are to receive which treatment, it is important to use random assignment so we can remove bias from our study. It is not appropriate to let the subjects decide which treatment they wish to receive. In an experiment, we wish to look at the differences between the treatment groups. There are usually other things that have an effect on the changes we witness in the subjects from the beginning to the end of the experiment. These outside Sources do cause bias. However, by using random assignment, all of the groups are biased in the same way. Understanding that it is the differences between the groups that interest us, the biases cancel each other out and allow us to more clearly identify the differences that are due to the treatment. (see page 338) Confounding (page 344-345) In many experiments, there is more than one factor being measured. When the levels of one factor are associated with the Ievelsof another factor, we say that these two factors are confounded. In our experiment there is only one factor (Brand A or Brand B medication), so confounding will not be a concern. Suppose we modify our experiment so that, in addition to taking a daily medication, each subject was also put on an exercise program. Half of the subjects are asked to walk 2 miles/day and the other half are asked to swim 2 miles/day. Our experiment now has 2 factors (type of medication and type of exercise). Including the piacebo group, we have 4 levels for each factor (0 mg/day, 5 mg/day, 10 mg/day, and 15 mg/day). So, we now have 42 = 16 experimental treatments. It is almost certain that, when comparing differences between the groups at the end of the study, the effect of combining the two factors will be confounded (mixed) with each other. This does not mean that a study of this nature is inappropriate. Just that it will be very difficuit to differentiate how much of the differences in blood pressure that the subjects experience (if any) are due to the medication and how much is due to the exercise program. Diagrams (pages 334-335) An experiment is carried out over time with specific actions occurring in a specific order. Diagramming the experiment provides a picture for organizing the experiment. Back to our experiment of giving the subjects either Brand A or Brand B medication (with no exercise program). Let’s further use the basic experiment first introduced, where each subject got the same amount of medication (10 mg/day). Diagram the experiment. Be sure to read about the Four Principles of Experimental Design at home (see pages 332-333). 1. Control — Variation from sample to sample’will always occur (called sampling variability). When designing an experiment, it is important to control the sources of variation other than the factor(s) we are testing. When the subjects are human beings, we accomplish this by treating all subjects alike. Controlling extraneous sources of variation reduces the variability of the responses, making it ‘easier to detect differences among the treatment groups. Randomize — Randomization allows us to equalize the effects of unknown or uncontrollable sources of variation. If the subjects in the experiment are not selected randomly, there is a high degree of potential for biased data H which leads to biased results. If the results are biased, you cannot use them to draw any useful conclusion about the underlying population. . Replicate — Since results vary based on the subjects in the study, a researcher should always repeat the study to identify and reduce the effects of this variation. Furthermore, after replication, if the experiment’s results areconsistent we have stronger evidence to support the study’s conclusions. Block — in many experiments, there is more than one factor being measured. When the levels of one factor are associated with the levels of another factor, we say the factors are confounded (pages 344- 345). Blocking is used to separate the confounded factors. If we group (block) individuals together and then randomize within each of these groups (blocks), we can remove much of the variability due to the confounding of other factors with the factor we are trying to measure. Blocking is the arrangement of experimental units into groups (blocks) that are similar to one another. Blocking reduces known but irrelevant sources of variation between units and thus allows greater precision in the estimation of the source of variation under study. Diagram the final version of the experiment developed throughout this lesson. This is the experiment that has 2 factors (medication & exercise program) and 4 levels (0 mg/day, 5 mg/day, 10 tug/day, and 15 mg/day). ...
View Full Document

Page1 / 4

Chapter13 - Chapter 13 — Experiments &...

This preview shows document pages 1 - 4. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online