METHODS OF SOCIAL RESEARCH

Kenneth Bailey

Free Press, 1978

note: some of the equations and tables are not browser friendly....

CHAPTER 1 - Stages of Social Research

1. Choosing the research problem and stating the hypothesis

2. Formulating the research design

3. Gathering the data

4. Coding the analyzing the data

5. Interpreting the results so as to test the hypothesis

CHAPTER 2 - Factors Afflicting Problem Selection

2. Researcher's values - Goffman vs. value free so state

3. Degree of reactivity inherent in methodology used

4. Methodology used including degree of proof required

5. Scope of the study

6. How time is treated

Cross-sectional vs. latitudinal studies

CHAPTER 3 - Constructing Social Explanations

Descriptive (to describe) vs. explanatory prediction study. Def; concepts that can take on more than one value along a continuum are called variables. If only one possible value it's a constant.

Theory construction

1. formulate basic concepts

2. write propositions (subtypes include: hypothesis, empirical generalization, axioms, postulates, and theorems)

(a) one variable - invariate

(b) two variables - bivariate

(c) three or more - multivariate (should try to make into bivariate)

A hypothesis is a testable statement; an empirical generalization is a relationship that represents an exercise in induction. One observes a relationship and then generalizes to a broader category.

Axiomatic theory: Postulates, axioms, theorems -- it takes the form of the deductive syllogism.

Bivariate relationships: Positive or negative, strength of relationship, symmetrical, asymmetrical (independent and dependent variables), linear or curvilinear, spurious or involves an intervening variable.

Generally the dependent variable is the one we wish to explain and the independent variable is the hypothesized explanation. A suppressor variable suppresses the relationship by being positively correlated with one of the variables and negatively correlated with the other.

A ¬ C -->B<--A C--> B

Spurious Intervening

Blalock (68) = empirical phenomena must be directly amenable to detection and monitoring by senses of observations, touch, hearing, smell.

Three basic hypothesis testing process: classical, grounded, operational.

most complete ³ Conceptual level x r1 y

³ ¯ ¯

high possib. ³ r2 r3

of measurement ³ Empirical level ¯ ¯

error ³ x1 r11 y1

x1 and y1 are called indicators (measures, scales, indices) of the concept r2 and r3 are generally called epistemic relationships and are usually untestable and therefore assumed.

Grounded theory: that developed from data.

1) enter the field work phase without a hypothesis

2) describe what happens

3) formulate explanations on basis of observations

measurement ³ Conceptual level x r1 y

error reduced ³ ­ ­

limits ³ r2 r3

generality ³ Empirical level ­ ­

x1 r11 y1

Operationalism: P.W. Bridgeman defined as we mean by any concept nothing more than a set of operations.

measurement ³

error absent ³ Conceptual level x r1=r11 y

by dfn ³ ô ô

can't abstract ³ Empirical level x1 y1

CHAPTER 4 - Measurement

Nominal measurement is a classification system, must be at least two categories and they must be distinct, mutually exclusive, and exhaustive (a category for every case)

Ordinal ranking

Interval - how many units of difference between

Ratio - a true zero point

Validity

face validity: determining if the instrument arrives at the concept adequately

criterion validity: involves multiple measurement of the same concept

construct validity: replace test sets and run through construct to see if valid for both indices.

Conceptual level x1 x1 x1 « x2

¯ ÷ ø ¯ ¯

Empirical level x11 x11 « x111 x11 « x21

faceV. criterionV. or ö÷

x111

constructV.

Internal validity asks whether a difference exists at all in any given comparison.

External validity is the problem of interpreting the difference, to generalize the results how far.

Assessing reliability (1) by use of alternate or parallel forms of the same measure used simultaneously; or (2) repeated application methods; or (3) split half method whereby researcher constructs a single instrument containing twice as many items as needed.

CHAPTER 5 - Survey Sampling

Objects of study called units of analysis {x, x, ...} - population or universe -- where x is called a sampling element of sampling unit a sampling frame

1968 Nixon % called by Harris (41%) and Gallop (43%) actual (42.9%) samples. Size 2000 of 73 million.

Random sample - develop sampling frame of adequate size to pull randomly - sampling units or elements.

Simple random sampling - sampling without replacement. Means ignore subsequent number of one already drawn.

Appendix A - random tables explained pg. 76

If the researcher can find no evidence of biased ordering there is little recourse but to assume random order.

Systematic sampling easier if one doesn't have a sampling frame completed. Best if sampling frame is already randomized [considered expensive]

Two assumptions for systematic sampling

1 that elements appear in random order with regard to characteristics of interest

2 sufficient number (K) over time (T)

Random interviewing street corner pick one of first x and then every x+x or x+10, etc.

Stratified random sampling adv. smaller sampling size. Mendenhall, Ott, Sheaffer (1971) defn: separating population elements into non-overlapping groups, called strata, and then selecting a simple or random sampling from each strata.

Dfn: Cluster sample (sometimes called area sampling), is generally used when it is impossible or impractical to construct a sampling frame in which the sampling units are the sampling elements themselves.

Disadvantage lack of control because multiple samples.

Nonprobability samples: ____vience, quota, dimensional (quota across universe), purposive (use of research and knowledge), snowball uses interview to find requisite characteristics then uses them to name others

Mean of sampling distribution µ mean for single sample

µ = Sn xi¤N

Variance adds squares of deviances so as not to have signs cancel each other

Variance = Sni=1(xi-x)2¤N = symbolized s2 for the population or universe and S2 for the sample.

Therefore standard deviation s or S = Ö(x-)2¤N

The greater the S or s the greater the sample size must be. This is a sign of heterogeneity. If s is small, can perhaps reduce.

Standard error S.E. = (s¤ÖN)(Ö1-f¤1), where f is the sampling fraction # in sample . Note that as f approaches 1, S.E. &reg; 0,

# in population

f approaches 1 as population ¥ Small f would mean small N so large S.E.

CHAPTER 6 - Questionnaire Construction

Longitudinal (time) studies: panel studies (same respondents), (topic) trend studies

Key word is relevance: avoid double barrelled, ambiguous, leading

closed-ended limited response

open-ended

Format - get quick, easy, closed questions over so as to gather data.

Mailed: cover letter, instructions, pretesting.

CHAPTER 7 - Mailed Questionnaires: good & bad

CHAPTER 8 - Interview Studies

Flexibility, response rate, nonverbal behavior, control over environment, question order, greater complexity, question order, open-endedness.

Disadvantages: cost, time, less anonymity, bias.

Use prepared questions.

This is an attempt to record how personal composure and job conditions are related. In particular I am studying individual reactions to pressure situations. Because of the nature of my occupation - firefighting - I am able to witness close encounters with stress. Being actively engaged in my job responsibilities I am unable to record as my observations occur. Also I am forced t keep my status as a researcher a secret from those I work with. This is because of their dislike of observation and to prevent contamination. For these reasons I have decided to do a field history record as soon as possible after the incident.

An incident begins with a toned alarm. The tone is loud and calls all hands to alert attention. As incidents present the stress situation I have decided to do an incident-by-incident field history rather than record daily.

PART 3 - NONSURVEY DATA COLLECTION

CHAPTER 9 - Experiments

Control over causal variable (the experimental stimulus - often a movie). Can measure values before and after accurately.

1 Establishing causality: debatable if one can prove causality empirically

2 Control

3 Longitudinal analysis

1 Artificial environment

2 Experimentor effect

3 Lack of control soc and people make control difficult

4 Sample size

Experimenter must show closure.

Process: State hypothesis; measure dependent; introduce independent; remeasure dependent

Do this to separate groups leaving out independent and show causality.

202 Solomon 3 group: only by adding a third control group, which as neither pretest nor test-stimulus effect, can we isolate extraneous effects. This is actually to experiment groups with one control group.

Factorial design: good for using variance. Cross checks variables involved.

Latin square design: presents as many independent variables (experimental conditions) as there are subjects, but variables presented in unique and different order for each subject.

Assignment of subjects:

Simple matching (find matching pairs and assign exp. & c)

Frequency dist: makes groups similar on average value of one variable.

Randomization.

Research outside of lab: Asch's study of conformity where whole group set up on length of lines with one real subject.

Semi-experimental designs: studied in natural environment.

Ex post facto experiment: determining causal factors from survey data.

Uncontrolled experiment.

Field experiments: the robber and the bystander: cases of beer.

Validity & Reliability:

Valid to extent we can measure effect of independent on dependent.

CHAPTER 10 - Observation