Read Cozby Chapter 9
Conducting Experiments
Setting the stage
Informed consent (as discussed earlier) is important for carryingout prior to conducting research. Don't give so much information that they simply confirm your hypothesis.
Give general information about the study with 'deception' and be sure to prepare for debriefing and the alleviation of any adverse effects of participating.
Types of Independent Variable Manipulations
Straightforward makes use of easy and clear alterations of variables such as written vs. oral vs. pictorial. E.g., memory study written (printed or computer generated) vs. verbal list of words (read or taped); Hot vs. Cold attribute in personality descriptions; anti-smoking information text vs. pictures of diseased lungs.
Staged Manipulations (event manipulations)
Why? 1) to produce a psychological state, expectations, emotions, thoughts, behaviours-gives context for interpretation. 2) simulate real world.
E.g., frustration (anger) vs. happy mood & adrenaline; work environment (accident, learning, shocking) often use confederates to set stage (lines length judgments & conformity).
Other Concerns
Strength of variables- two or more levels, try to make them as strong as possible to observe effect, but stay realistic (i.e., crowding + ethics). May want to know the minimal strength necessary to elicit response.
Cost also comes into the picture, what can you afford (materially or time wise)? In terms of...
Ethics - must always be a concern in research
Measuring Dependent Variables -
What are you looking for?
Self Report have people tell you what their attitudes or feelings are through questionnaires or interviews.
Behavioural recording of type, reaction times, rates, or duration can easily be made.
Physiological recording of functioning such as GSR, EMG, EEG, EOG, Blood Pressure, Heart Rate, ...
Multiple measures are often the best way to go. Use similar or contrasting methods to study various ways to express or measure the same thing.
E.g., Anger-self report, behavioural & physiological while performing discriminant validation on the operational constructs from several theories.
Construct validation by comparing various measures and the theories that predict their relationships.
Other
Concerns
Cost
of recording and acquiring data
Ethics of making such recordings
Additional
controls
Demand
Characteristics - any features / aspects of the experiment that
might inform subjects of the true purpose of the study and alter their
responses.
Expectations may have impact (i.e., studying cheating-no honest answer due to social desirability.
Use filler or distractor items to hide real questions in response to expected response bias.
Placebo groups are also used to control for test expectations only.
Experimenter bias also may creep in where subjects are treated differently w.r.t. the condition they are in; e.g., smile at or greet more warmly those that you know will get adrenaline.
Also may arrive in form of
biased recording or interpretation of protocols
based upon experimenter expectations.
(Clever Hans)
Solutions?
Train
experimenters
to act consistently
Automated or standardised presentation of the conditions (computer, random lists)
Single And Double-blind studies where neither experimenter nor participant knows what condition.
Debugging
the study
Making
proposals to identify the background context, purpose, procedures,
participants, measures and interpretations.
Getting
feedback on the intentional and practical implications of the proposed
research. Include: Literature review, methods, hypotheses,
ethics.
Pilot Studies to try out the materials and your ability to carryout the procedure.
Manipulation
checks are done to find evidence for construct
validity. (I.e., anxiety - does it actually measure anxiety?
Have self-reports on the manipulation as to whether or not it worked.
Debriefing
- discuss the ethical and educational implications. Ask whether or not
they know what it was about and inform them
of the actual intent.
Provide information for follow-up or support
to overcome stress or discomfort.
Read
Cozby Chapter 10
Complex
Experimental Designs
Increasing the number of levels of IV
Simplest designs use only two levels of Independent Variable (Treatment & Control). We often assume that there is a liner or monotonic relationship between the 'variables' we measure and record.
By examining multiple levels of specific variables we can better understand its relationship with other variables.
Increasing number of IVs - Factorial Designs
Factorial
designs are used to examine more than one independent variable (factor)
with two or more levels each.
E.g., 2 x 2 design using gender and speed of processing (fast or slow)
on word recognition.
Can have all sorts of designs
with various levels. E.g.,A study on LSD and Social Interaction 2x3x4 where
there are two levels of Var1 (setting-club, party) and three on Var2 (dosage-500,
1000, 1500) and four on Var3 (music-hiphop, techno, psychedelic, grunge)
Interpretation of Factorial designs usually hinges upon the main and interaction effects. These are calculated statistically where deviation scores (distance of each person from mean) are used. Take the variance (sum of squared deviations) for group versus overall variance; is it significant?
Assuming that each factor has some kind of causal potency the 'effect' of the factors alone are estimated along with the 'effects' of the two or more factors combined. E.g., gender only as an influence on word recognition or the combined influence of gender and reading speed on word recognition.
Factor
Designs with and without manipulated IVs
Manipulated
independent variables are those that we put trust in as having causal
influence on the DVs because we attempt to control for all other possible
influences.
Other
Independent Variables are used to compare characteristics of participants
or the social/moral (everyday) groups they represent.
These "Subject
Variables" constitute a number of "IV"s in psychological research.
Gender, age, ethnic group, language group, delinquents, working poor, homeless,
etc.
In essence these groups are
artificial where they often represent a fabricated group based upon selection
(sampling) procedures of the researchers. Causal attributions are made
based upon the probabilities of finding the observed differences in sums
of squared deviation scores.
(see chapter
13 later).
Assignment
Procedures and Factorial Designs
Mixed
factorial designs stand in contrast to the completely between (independent)
groups design and the within subjects (repeated measures)
design.
Thus mixed factorial designs
make use of both types of design in the same study
(between-within designs).
Thus subjects get both conditions of one variable and one condition of
the other in a 2x2 design.
Savings in number of subject 20 vs. 40 (indep.) and 10 (repeat), but allows some independence of groups on factors that might not work well on repeated measures. Middle road.
Selecting
an analysis for your data:
Relationships
among variables
Discrete
and Continuous variables and the families of statistics that can
be performed rather diverse; tests are defined by types of variables.
Nominal and Ordinal scales are discrete as they represent categorical data. Either/or - binomial, this that or the other thing - trinomial, integration assimilation separation and marginalisation - polynomial.
Interval and Ratio scales are considered to be continuous where proportions or fractions are possible and meaningful. On a continuum there is not either/or, black or white, but many shades of grey in between.
Single variables are like single colours, each can be mixed in larger or smaller quantities with the others to make an infinite variety of colours representing all of the human psychological traits and features.
Adding multiple levels to variables gives a range of intensities (brightness) for any given colour. Along with the addition of colour hues the addition of variables gives a richer image of the phenomenon.
The drawback of having multiple levels and multiple variables is the alignment of their scales while trying to make meaningful comparisons among them.
How different scales can match up
Last week in Lab we considered what happened with nominal and ratio variables when asked to perform certain statistics.
Nominal data can be described clearly using graph or category frequencies, however they don't perform well when asked for means and deviation scores. (note see SPSS-Frequencies along with graphs)
On the other hand, interval & ratio data do perform well when asked for means and deviations scores. Frequencies also can be revealing.
This week we will look at what kinds of tests can be done with certain kinds of variables.
(See chapter 13 page 236 for a table)
Nominal
Data - Chi Square test
One independent variable of
two groups or more.
Use the expected frequencies of the specific categories vs. the observed frequencies. Thus using the sampling distribution of Chi2 to examine the probability of the observations occurring simply due to chance. If likely, then not significant. Null hypothesis assumes that all groups are equal in number of people of each type.
However, if the likelihood is very small that these results occurred by chance then the pattern of observed frequencies is said to be statistically significant.
Nominal data can also be used as independent (SV)s or blocking variables. E.g., gender, ethnicity, acculturative style, ego identity style, identity...
Interval or Ratio data (DVs) for single variables of two independent groups can use t-test of significance.
For more than two groups (levels) use one-way ANOVA for single variable. Two- or three-way ANVOA is done for two or three factors being considered (or four factor, ....)
Chapter
11 ( developmental studies are relevant for studying
-particularly the comparison between longitudinal and cross-sectional studies).
Next Lecture (week 13): Nov 26th
Chapter 12 Descriptive Statistics and Correlation
Week 13 Lab 8 - Correlation and Regression
Then: Dec 3rd will cover
Week
14: Chapter 13 Inferences
Lab Project Presentations
December 5th
Take
home final exam Due December 10th
or on-line
final December 11th