Brief Lecture Notes 2 (revised September 5, 2006)
Important note: The department assessment activity (required of all students) will cover basic aspects of descriptive statistics. See here for an overview, as well as text pages 64-66.
The branch of philosophy known as epistemology deals with the general question of how human beings come to know anything, or how we determine what is true and valid. This isn't a philosophy class, but since Unit 2 deals with the scientific method (which is, after all, one specific way of trying to find out things), it's helpful to begin with a wider context.
According to Colin Chapman, there are six general ways of knowing that (in his view) cover the waterfront:
1. Induction means generalizing from direct experience or direct sensory observation.
2. Deduction means logical (i.e., syllogistic) reasoning from abstract premises or presuppositions.
(Taken together, the use of these two ways of knowing in combination is sometimes known as retroduction.)
3. Authority means relying on the word of someone you have reason to regard as trustworthy or expert in a given domain of knowledge or universe of discourse. (Human authorities necessarily have expertise only within a given circle of knowledge, and do not retain that mantle of expertise when they speak of matters outside that circle.)
4. Intuition, for Chapman, covers all subjective forms of knowing including hunches, guesses, gut feelings, insights, and the like. In general, intuitive knowledge is difficult or impossible to justify in verbal terms (we'll learn why in Unit 3).
5. Consensus means arbitrary social agreement or social convention. The emphasis is on the arbitrary (or purely utilitarian) nature of a group's agreement to view things in a certain way. (Unexamined social consensus may sometimes be mistaken for intuitive knowing.)
6. Existential commitment (like Pascal's famous wager) means choosing to act with more certainty than can be justified by means of the other ways of knowing because of the costs of not doing so or because of the demands of living that require decision and action in the face of ambiguity or less than full certainty.
For instance, here is how each way of knowing could be used to answer the question, "How do you know it is raining?"
1. Induction: I looked out the window and saw for myself.
2. Deduction: I hear a pattering sound on the roof. The only phenomenon that can produce that pattering sound is rain. Therefore I conclude that it is raining.
3. Authority: The TV meterologist says it is raining.
4. Intuition: I just have a hunch that it's raining. I can't tell you how I know, but I do.
5. Consensus (most difficult to apply to this example): My friends and I have agreed that whenever it is cloudy and there is discernable moisture on the windowpane, we will define that as "raining".
6. Existential commitment: I don't know for sure if it's raining or not, though rain strikes me as likely; but I'm going to take an umbrella with me anyway, because if it is raining and I am without an umbrella I will get drenched, but if it isn't raining and I have an umbrella I am simply slightly inconvenienced or embarrassed.
Each of the six ways of knowing has some liabilities or problems associated with its use, especially if it is taken to excess or utilized in isolation. (To go into detail about that would require that this become a philosophy class.) Thus, the best approach may be to utilize all six of them in a complementary, "checks and balances" fashion... and, indeed, all human beings (including scientists engaged in their professional roles) do make frequent use of all six ways of knowing on a regular basis. Thus, even though the scientific method (our formal topic for this unit) makes primary and systematic use of the first two ways of knowing, induction and deduction, don't ever let anyone tell you that scientists don't use all six ways of knowing in their scientific work. They do.
For a question to qualify as scientifically testable, it must be both empirical (subject to verification by means of sensory observation) and replicable (capable of repeated observation or testing). See text pages 39-40, though I disagree with the textbook authors about the nature of scientific objectivity. My own view is as follows:
In my view, science cannot answer all questions because it deals with matters of repeatable observation. It can only treat data that is empirical and replicable. For this reason, science cannot address ethical questions (since science is about what is -- hence, at least in part, actually or theoretically observable -- while ethics deal with what should be). Similarly, historical questions are technically outside the province of science since historical events are, by definition, unique and unrepeatable. This does not necessarily imply that ethics and history are "subjective" fields in contrast to the supposed "objectivity" of the pure scientist, as Mortimer Adler (to take one example) notes in detail.
Of course, in practice scientists often engage in "leaps of logic" that extrapolate rather far beyond the extant data, and to note this is not necessarily to be critical. The question of how far such "quantum leaps" in reasoning can rationally go before one is justified in accusing a given theorist of engaging in wish fulfillment is a difficult and technical one, although wherever the bar is placed, it should be applied in a philosophically neutral way to the extent that such an enteprise is possible.
The above argument is in rather sharp contrast to the prevailing view in our culture that matters of "fact" (which presumably can be derived only by means of science) and matters of "value" (which derive, it is argued, from the use of other ways of knowing that are inherently "subjective") occupy wholly different, incommensurable and nonoverlapping, realms. My contrasting contention is that, to the extent that human beings can be approximately objective (or at least balanced and able to correct substantially for their own biases), science provides one means, but by no means the only means, of gaining objective knowledge. In other words, not all truth is scientific truth.
Four goals or aims of science are description, prediction, explanation, and control (see text page 40). Description means stating in precise terms what a phenomenon is like, what is all about, what its key features are (answering "what" questions). Prediction means stating when (that is, under what conditions or circumstances) the phenomenon is likely to occur, based on an observed association or correlation between that phenomenon and some other event or condition (answering "when" questions). Explanation means identifying a causal mechanism by which the relationship between the two events above can be understood (answering "why" questions); note that while correlations are observable, cause-effect relationships are not and must be inferred. (Hence science requires the use of both induction and deduction.) Control means modifying the probability of, or the characteristics of, the phenomenon, based on a real-world application of the explanatory model or theory (answering "how" questions). Some sciences obviously must stop with explanation, control being humanly impossible or meaningless; can you think of an example? See text pages 41-42 for more on the relationship between theory and data.
As noted above, a correlation is an observed association between variables (see also text pages 65-66). Correlations can be positive (increases in one variable are associated with increases in the other) or negative (increases in one variable are associated with decreases in the other). Correlations have both a direction (positive or negative, as above) and a magnitude (the extent to which one variable can be reliably predicted based on knowledge of the other). Correlational relationships hold true only on the average (for an entire sample or population of observations), not necessarily for any single given instance. For instance, while in general height and weight are positively correlated (tall people on average weigh more than short people), it's possible for a single human being to be 3' 4" in height, 750 pounds in weight. Such a single observation, which is highly discrepant from the general population trend, is called an outlier.
Correlation does not necessarily imply causation; just because two variables are correlated does not mean that one of them is a direct cause of the other. When two variables are correlated in the absence of a causal relationship between them, this is called the third-variable problem (see text page 45). For instance, among a group of mixed-age children, shoe size is positively correlated with reading ability (kids with bigger feet tend on average to read better), but only because both shoe size and reading ability increase with age; the correlation disappears if children of a single age are examined, and is thus an illusory correlation or artifact. Neither variable causes the other (reading does not make your feet swell, nor will getting silicone foot implants make you more literate).
Research begins with a general question of interest which must be turned into a testable research hypothesis. (See text pages 39-42.) Not all questions are hypotheses; some of the most important questions that can be asked, such as "What is the meaning of life?" cannot be turned into scientifically testable hypotheses, which is another way of saying that science cannot answer all questions (because the inductive and deductive ways of knowing are not self-contained, all six of them being necessary). To qualify as a testable hypothesis, a question must (a) state a presumed cause-effect relationship (b) between at least two variables (c) that can be operationally defined, that is, a precise statement can be made of how, in objective and observable terms, the variable is to be measured or assessed. Obviously, some variables are easier to operationalize than others; it is easy to determine a person's height, more difficult to determine his or her integrity.
Be sure that you can distinguish between variables and levels of a single variable. For instance, height is a variable; being 5' 8" in height is a specific value of that variable. Variables, of course, must be able to take on more than one level or value (that is, they must be able to vary); a "variable" with only one possible value is no variable at all, but a constant.
Research designs (see text pages 42-52) can be characterized as experimental, correlational, or descriptive (see below for definitions), and each can be conducted in either a laboratory (artificial) or field (real-life) setting, yielding 3 x 2 = 6 possible designs.
Experiments are defined by the direct manipulation of one or more independent variables, as well as the quantitative measurement of one or more dependent variables. An independent variable is directly controlled by the experimenter (so that, for instance, s/he could randomly assign a research subject to whichever treatment condition or level of the variable s/he chose, with no pre-existing constraints). Experimental designs can be either between-subjects (each subject receives one and only one treatment condition) or within-subjects (all subjects receive all treatment conditions in turn). In the simplest between-subjects design, there are two groups: the experimental group (which receives some treatment or intervention) and the control group (which receives no treatment, or more likely, a placebo -- something that has the outward appearance of a genuine intervention or treatment but which really is expected to have no effect). The point of the experiment is to test the hypothesis that observed differences in the dependent variable are caused by induced changes in the independent variable. To make sure that this is a valid conclusion, the groups must be alike (and treated alike) in all other respects to avoid the existence of a confound -- an alternative explanation for the observed differences on the dependent variable. Confounds cannot be eliminated entirely, but they are kept to a minimum in a well designed experiment. Even absent any confounds, it is necessary to rule out the particular kind of alternative explanation known as the null hypothesis (the notion that the groups differed on the dependent variable merely by chance), which is accomplished probabilistically using statistical analysis.
If there is no true independent variable, the design is not an experimental one; if one or more variables are observed, but none are manipulated, it is a correlational study. (If there is not even any systematic numerical measurement of variables, which usually means no specifically testable research hypothesis, it is a descriptive study.) Three reasons why a true experiment may not be conducted are that (a) it is literally impossible to manipulate the causal variable, (b) it is possible but unethical, (c) it is possible and ethical, but not cost effective.
For instance: to test the hypothesis that women are more generous than men, I approach 50 women and 50 men at the Wausau Center Mall, asking each for some money. I record the amount of money offered as the outcome measure (if I am ethical, I then give the money back). This looks like an experiment but is not, because what appears to be the independent variable (gender) is not really being manipulated, only observed (measured). To manipulate the variable would mean randomly assigning subjects to genders (heads, you're now male; tails, you're female), which not only requires expensive gender reassignment surgery in many cases, but which might generate some understandable resistance on the part of some subjects.
In deciding on a lab versus a field setting, one needs to keep in mind the control-realism tradeoff. Lab settings maximize control (over potential confounding variables), but are often notably unrealistic. Field settings maximize realism (are, by definition, closest to life as it is actually lived), but are often fraught with potential confounds. Since both control and realism are desirable properties of research, yet anything that is done to increase one will generally decrease the other (the two variables are negatively correlated!), one has to prioritize and then maintain a delicate balancing act in selecting a research design and setting.
Ethical issues in research include confidentiality, informed consent, and questions relating to the proper use (if any) of deception in research. See text pages 60-61.
For a helpful practice problem (similar to what you'll find on the first exam!) relating to research methods and designs, click here. Parts 2 and 5 of that question weren't covered in our class, so relax... you don't have to know that material. But parts 1, 3, and 4 should be familiar, old friends.
Study Guide 2
1. Define, and compare/contrast, the six ways of knowing. Be able to give examples of each.
2. What are four aims or goals of science? Explain, and give an example of, each.
3. What three characteristics must a research question have to qualify as a testable hypothesis? Give some examples of questions that do, or do not, qualify as formal research hypotheses.
4. What is a correlation? How do positive and negative correlations differ? Give some examples of each. What does it mean to state that correlation does not necessarily imply causation? That while correlation can be observed, causation must be inferred? What is the third-variable problem? Give an example.
5. How do experimental, correlational, and descriptive research differ? Lab versus field research? Between-subjects versus within-subjects research? Be able to give examples of the various research designs or to recognize (classify) examples. What are the relative advantages and disadvantages of the various design alternatives?
6. How do independent and dependent variables differ? How do manipulation and measurement differ? What is a confound, and how can confounds be minimized? What is a placebo, and what is their role in research? How do experimental and control groups differ? What is the null hypothesis, and how can it be addressed or handled?
7. State and explain three reasons why it is not always possible to conduct a true experiment, and give examples of each.
8. Summarize three ethical issues in research and how they might be addressed.