1. P1) X has properties a, b, c, and z. P2) Y also has properties a, b, and c. C) By analogy, Y has property z. X: primary analogue Y: secondary analogue A difference between X and Y is a disanalogy.

2. P1) X has properties a, b, c, and z. P2) Y also has properties a, b, and c. C) By analogy, Y has property z. X: primary analogue Y: secondary analogue Analogical arguments are strong when there is an important way that a, b, and c are connected to z. If you find an additional primary analogue (let’s call it W) where a, b, and c are present, but not z, that casts doubt on the connection. Such cases are called counteranalogies.

3. P1) X has properties a, b, c, and z. P2) Y also has properties a, b, and c. P3) W has properties a, b, and c, but does not have z. C) By analogy, Y has property z. X: primary analogue Y: secondary analogue Usually, increasing the number of primary analogues increases the strength of the conclusion. But if you have additional primary analogues that are counteranalogies, they weaken the strength of the inference.

4.  Mill’s methods closely resemble certain scientific methods related to causal connections and correlations.

5.  For example, the Method of Difference is virtually identical to the method of controlled experimentation used in such fields as biology, pharmacology and psychology.

6. Controlled experiments induce a condition in which you can narrow down the source of an effect… Possible Causes Occurrence ACDEFG Phenomenon 1 ✓✓✓✓✓✓✓ 2 ✓✓✓✓✓ --

7.  Controlled experiments involve two groups of subjects: the experimental group and the control group.  The control group and experimental group are ideally subject to the exact same conditions except the experimental group receives the treatment when the control group does not.

8.  The first controlled clinical trial was performed in 1747 by James Lind, who used it to investigate whether scurvy could be cured with citrus fruits  His knowledge of how to avoid scurvy, and also of how to prevent the spread of typhus has been credited with allowing the British Navy to get a competitive advantage over the French  Lind noticed that typhus disappeared in parts of his hospital where patients were given clean clothes and sheets and bathed, but was present in parts of the hospital without hygiene measures

9.  Controlled experiments involve two groups of subjects: the experimental group and the control group.  The control group and experimental group are ideally subject to the exact same conditions except the experimental group receives the treatment when the control group does not.

10. For example: ◦ You want to know if an experimental ADHD treatment is effective ◦ Select 100 children with ADHD, and assign 50 to receive the treatment and 50 to receive a placebo ◦ Check rates of “problem” behaviors the treatment is supposed to help with

11. Controlled experiments are important because they allow you to rule out sources of error. For example, let’s say you want to see if a flu treatment is effective. If you only test using one group of people, and check to see if they improve, you might mistake the natural progression of healing for effectiveness on the part of the medication.

12.  You need to be able to show that the changes you observe can be legitimately attributed to the treatment.  In that case, you would want a control group of people who also have the flu, but don’t receive any treatment.  If the two groups recover identically, it’s a sign that the treatment isn’t effective.

13.  However, such an inference would still be vulnerable to error. Another way to reduce the possibility of error would be to randomize the groups.  Randomization prevents selection bias ◦ If you don’t randomly assign people to your two groups, it could be the case that subjects with a certain cluster of traits end up in either the treatment or control group. ◦ The effects you observe might then be only a result of the different traits that existed to begin with.

14.  CS Peirce argued for the importance of randomization-based inference  He showed how randomization is essential to allow for strong inferences in experimental and survey research  Introduced the first RCT trials in psychology (also first instances of blinding)

15.  The method of difference is useful when you have a pre-selected factor to look at  If you’re trying to look at a class of possible predictors, and determine which one is responsible for a kind of outcome, you will want to use the joint method of agreement and difference

16.  The salient difference between Mill’s method of difference and his joint method is this: ◦ The difference method determines whether a preselected condition is the cause of a phenomenon. ◦ The joint method takes a preselected class of possible conditions, and applies rules for finding sufficient conditions and necessary conditions to hopefully reduce the class of conditions to just one.

17.  E.g., retrospective nutrition study ◦ Nutritionist wants to determine how different supplements could affect artery disease ◦ She asks participants to report what they ingested over the last few years ◦ She checks their health status: artery disease or no artery disease: ◦ She analyzes the data and determines which, if any, supplements are present in cases where people have artery disease versus ones where they don’t

18.  The correlational method in the social sciences closely follows the method of Concomitant Variation.  A correlation is a type of association between variables.

19. Example:  A professor randomly selects 100 students, pulls their GPAs, and makes them take an IQ test  He finds that students with higher GPAs generally have higher IQs and those with lower GPAs generally have lower IQ’s  This would indicate a positive correlation ◦ The degree of correlation is calculated in terms of a correlation coefficient

20. Example:  A professor randomly selects 100 students, pulls their GPAs, and makes them take an IQ test  He finds that students with higher GPAs generally have higher IQs and those with lower GPAs generally have lower IQ’s  This would indicate a positive correlation ◦ The degree of correlation is calculated in terms of a correlation coefficient

21. Example:

22.  If an increase in one variable tends to be associated with an increase in the other, there is a positive correlation. ◦ e.g.: Height and weight are positively correlated.  On average, as height increases, so does weight.  If an increase in one variable tends to be associated with a decrease in the other, then there is a negative correlation. ◦ e.g.: Car weight and mpg are negatively correlated.  On average, the heavier a car is, the less mpg it gets.

23. Here we need to know: what other variables were looked out? Is there a possibility of systematic bias?

24. Our ability to make scientific inferences based on experimental studies is largely dependent on the degree to which we have scientific control. Scientific control is seen in controlled experiments in both random assignment and the use of control groups, but there are other forms of scientific control…

25. Key errors to avoid in scientific research: - Mistaking chance or noise effects as representative of a phenomenon - Specifying the wrong value for a parameter - Using an inappropriate experimental design

26.  Blinding ◦ Participants do not know which group they are in  Use of placebos (relation to method of difference)  Prevents results that are biased by patients acting to fulfill expectations  Double-blinding ◦ As above, except neither participants nor experimenters know which group a participant has been assigned to during the course of the experiment itself  Prevents experimenter expectation effects

27. Done properly, scientific control can all but eliminate alternative explanations of experimental results.