Mottos/Mantras/Maxims/Take-Home Lessons From Chapter 7

  1. Descriptive research may start with a question about a single variable (i.e., "how many people do ____?" or "how often does ___ happen?" ) but usually goes beyond looking at single variables to looking at correlations among variables (i.e.,  "When and where is the behavior more likely to occur?" and "Who is more likely to do the behavior?). Thus, descriptive research is often called correlational research
  2. Descriptive/correlational  research can answer "what?", "where?",  and "when?" questions, but not "why?" questions.
  3. Correlational research cannot answer "why" questions because correlational research can tell you that variables are related, but not how they are causally related. That is, knowing that two variables is correlated does not tell you which variable is the cause and which is the effect (the "chicken and the egg" problem) --or whether both variables are effects of some third variable (e.g., ice cream consumption doesn't cause sunburns and sunburns don't increase ice cream consumption). In other words, correlational research does not have internal validity.
  4. Correlational evidence, like circumstantial evidence, may suggest that a variable is a cause, but experimental research would need to be done to confirm that the variable really was the cause.
  5. Description aids prediction; description is how Netflix and Amazon are able to predict what you might like.
  6. To describe behavior and its correlates accurately, objective measurement is needed because perception is not objective, record-keeping is needed because memory is not accurate, and statistics are needed because people are not good at distinguishing between coincidences and  patterns.
  7. To make valid statements about constructs, we must have good measures of those constructs. Construct validity can be harmed by having a measure of a construct that is a poor reflection of the construct (as may happen in archival research), observer/coder bias (as may happen in observation and in content analysis), and participant bias (as may happen when participants are keenly aware of being observed).
  8. To generalize to a group, we must have data on a representative sample of that group. Usually, getting a representative sample means getting a large, random sample.
  9.  Ex post facto research often has high construct validity, but low external validity.
  10. Archival research, if it forces you to rely on a poor measure of your construct,  will have low construct validity. If, however, the measure is a good one, you may have good construct validity--particularly if the measure is nonreactive.
  11. If you have archival data on many participants over an extended time period, your archival study might have high external validity.
  12. Observation has traditionally been time consuming and difficult, but videotaping is making it easier.
  13. Research using tests often has good construct validity and can, depending on the representativeness of the people tested,  have good external validity.
  14. You can analyze correlational studies using tools like the t test and the F test that are used to analyze experimental data. However, using those tools to analyze correlational data does not give you the ability to draw cause-effect conclusions.
  15. You can often divide your participants into two groups by doing a median split--assigning all the participants who scored above the middle score to one group and all the participants who scored below the middle score to another group. Note, however, that by lumping everyone who scored above the median into one group, you have lost information: The person who originally scores 1 point above the median gets the same score as the person who scored 100 points above the median. Because you have lost information about differences between participants, you have lost power: the ability to find differences between conditions.
  16. There are different kinds of correlation coefficients: The one you use should depend on the level of measurement of your variables.
  17. Correlation coefficients can range from -1 to +1.
  18. A correlation coefficient of zero indicates that there was no relationship between your variables in your sample. The farther the correlation is from 0, the more related the variables.
  19. A positive correlation coefficient between two variables indicates that participants who were high on one variable were also high on the other variable and that  participants who scored low on one variable tended to also score low on the other variable.
  20. A negative correlation between two variables indicates that the variables were inversely related: Participants scoring high on one variable tended to score low on the other variable.
  21. A scatter plot of your data should show you whether you have a positive correlation coefficient (a line through the points slopes up), a near zero coefficient (a line through the points in your plot would be flat), or a negative correlation (a line through the points would slope downward). Such a plot may also alert you that the relationship between your variables may be nonlinear.
  22. To emphasize that the sign of a correlation coefficient has nothing to do with the strength of the relationship, realize that we get the best index of the strength of the relationship--the coefficient of determination--by squaring the correlation coefficient.
    Squaring the correlation coefficient means that you will always end up with a positive number. So, a  +.2 correlation coefficient and a -.2 correlation coefficient both give you a coefficient of determination of +.04.
  23. A correlation coefficient you get from a sample might be the result of a coincidence. To see if the relationship you found in the sample holds  in the population, you would do a statistical significance test.
  24. A significant correlation coefficient indicates that there is a relationship between the variables in the population. However, this relationship might be
    1. small
    2. only exist for the population from which you randomly sampled from
    3. nonexistent--if you did many statistical tests, some will be significant by chance alone.
  25. If there is a relationship between your variables, you may still fail to find a significant correlation between your variables because
    1. you didn't have enough participants
    2. you didn't have a sensitive enough measure
    3. your participants didn't vary enough on at least one of the variables. Two variables will not vary together (covary, correlate)if both variables aren't varying. In techincal terms, you have a restriction of range problem.
    4. you didn't have a linear relationship between your variables but you did have a nonlinear relationship.

Back to Chapter 7 Menu