research designs by grinnell & williams, 1990

Re: Research Designs
Fr: Grinnell, R. M. & Williams, M. (1990). Research in social work: A primer. Itasca, IL: F. E. Peacock, pp. 139-176.

3 Levels of Knowledge
“The best way to obtain data depends on what data we need. What data we need depends, in turn on how much is already known in our problem area” (p. 139).

KNOWLEDGE LEVEL - RESEARCH QUESTION (GEN.) - RESEARCH DESIGN (GEN.)
Very Little - General - Exploratory
A Little More - Specific - Descriptive
Much - Specific & Complex - Explanatory

3 Knowledge Levels & Corresponding Research Designs (p. 150)

Exploratory
1. One-Group Posttest Only Design
2. Longitudinal Case Study

Descriptive
3. Randomized One-Group Posttest Only Design
4. Randomized Cross-Section Survey Design
5. Randomized Longitudinal Survey Design
6. One-Group Pretest-Posttest Design
7. Comparison Group* Posttest Only Design
8. Comparison Group Pretest-Posttest Design
9. Interrupted Time Series Design

Explanatory
10. Classical Experiment
11. Randomized Control Group* Posttest Only Design

4 Conditions for True Experiment
1. Manipulate the Independent Variable (IV)
a. Basic Concepts: DV vs IV
Dependent Variable (DV) = what we want to change
Independent Variable (IV) = the thing that we hope will do the changing
Ex: Hypothesis: If: Uncle Fred enrolls in a quit smoking program (x) [IV],
then: he will quit smoking (Y) [DV].

b. Basic Concepts: EG vs CG
Experimental Group (EG): R O1 X O2
Control Group (CG): R O1 O2

2. Control for Intervening Variables (iv)
> “Control for” = identify & remove, or at least compensate for
> iv = something that interferes between the IV & DV; or something that
sneaks in when we are not looking & causes Y while we are still laboring
under the false impression that Y was caused by X.
Ex: One of the members of the EG brings a bottled lung of a deceased
smoker to one of the quit smoking sessions (not part of the regular
procedure).

3. Random Selection
= each member of the population has the same known probability of being
included in the sample, thereby ensuring representativeness leading to
generalizability of results

4. Random Assignment
= randomly assign the randomly selected respondents to 2 or more groups to
ensure equivalence* in terms of DV (ex: coughing behavior)
experimental group (EG)
Ex: coughers
control group (CG)
_________
*equivalent = only in terms of the meaningful variables directly relevant to the study

2 Basic Random Assignment Procedures
(1) Matched Pairs
Ex: measure level of depression (DV)
: rank order in descending sequence
: flip a coin or randomly select one of the 2 highest scores to be assigned to
either the EG or CG
: reverse the order of assignment for each pair, i.e., first pair assign highest
to EG, for 2nd pair to CG, etc.
: check by adding scores of both groups
Disadvantage: in odd numbered samples, a respondent or respondents could be
left out
NB: Not practical to match > 4 groups

(2) Matched Group
= each person is matched with a group, usually preformed (ex: existing
therapeutic group)
Ex: assign people to the “to-be-formed” comparison group by matching their
depression levels with the average depression level of the preformed EG
Random
Random
Assignment
Sampling Random

Assignment
Diagram of Random Sampling & Assignment

Research Designs
Rule in Choosing a Research Design: Principle of Parsimony = The simplest route to the object is always the best.

1. EXPLORATORY DESIGNS
Purpose: to merely gather data in an effort to find out “what's out there” (p. 170)

1.a. One-Group Posttest Only Design

FORMULA
X O1
Where: X = IV
Ex: Quit Smoking Program

O1 = 1st & only measurement of the DV
Ex: Success Rate: number of people who quit program

1.b. Longitudinal Case Study Design

FORMULA
X O1 O2 O3 ...
Where: X = IV
Ex: Quit Smoking Program

O1 = 1st measure of the DV
(1) Number of cigarettes smoked

O2 = 2nd measure of the DV
(2) Number of cigarettes smoked

O3 = 3rd measure of the DV
(3) Number of cigarettes smoked

2. DESCRIPTIVE DESIGNS
= with at least one & without at least one true experimental condition, usually lacking
either random assignment or control over intervening variables & sometimes both (p.
170)

2.a. Randomized One-Group Posttest-Only Design

FORMULA
R X O1
Where: R = Random selection from a population
ex: Pipe Smokers

X = IV
ex: Quit Smoking Program

O1 = 1st & only measurement of the DV
ex: Number of people who quit smoking

2.b. Randomized Cross-Sectional Survey Design

FORMULA
R O1
Where: R = Random selection from a population
ex:Random selection of people of Greater Pebble

O1 = 1st & only measure of DV
ex:1st survey of number of people who smoked pipes

2.c. Randomized Longitudinal Survey Design

FORMULA
R O1 O2 O3 ...
Where: R = Random selection from a population
ex: Random selection of people of Greater Pebble

O1 = 1st measure of the DV
(1) Number of cigarettes smoked

O2 = 2nd measure of the DV
(2) Number of cigarettes smoked

O3 = 3rd measure of the DV
(3) Number of cigarettes smoked

2.d. One-Group Pretest-Posttest Design

FORMULA
O1 X O2
O1 = 1st measurement of DV
ex: Number of people who smoked before the program

X = IV
ex: Quit Smoking Program

O2 = 2nd measure of the DV
Number of people who smoked after the program

2e. Comparison Group Posttest-Only Design

FORMULA
Experimental Group [EG]: X O1
Comparison Group [ComG] : O1
X = IV
EG: Quit Smoking Program
ComG: No Program

O1 = 1st & only measure of DV
ex: Number of people who quit smoking

2.f. Comparison Group Pretest-Posttest Design

FORMULA
EG: O1 X O2
ComG : O1 O2
where:
O1 = 1st measurement of DV
X = IV
O2 = 2nd measurement of DV

Ex:
GROUP
Pretest - Posttest = Difference
(O1) (O2) (O1 - O2)
EG 6 4 2
XomG 4 4 0

2.g. Interrupted Time Series Design
Purpose: to establish that whatever we want to change (Ex: Uncle Fred’s smoking behaviour) is not changing by itself before we introduce a treatment to change it [pretest] & to know the program results over time [posttest]; therefore, if posttests are significantly different from the pretests, we can be reasonably sure that our program is effective

FORMULA
O1 O2 O3 O4 X O5 O6 O7 O8 ...
O1 = 1st measurement of DV
O2 = 2nd measurement of DV
O3 = 3rd measurement of DV
O4 = 4th measurement of DV
X = IV
O5 = 5th measurement of DV
O6 = 6th measurement of DV
O7 = 7th measurement of DV
O8 = 8th measurement of DV

3. EXPLANATORY DESIGNS
Purpose: to explain things which have previously been discovered, i.e., a lot is already
known about the research topic & we are in a position to establish a particular IV x
causes a particular DV y (i.e., a testable hypothesis can be formulated on the basis of
previous studies) (p. 170).

3.a. Classical (True) Experiential Design or Pretest-Posttest Control Group Design
FORMULA
EG: R O1 X O2
CG: R O1 O2
Where: R = Random selection & assignment to group
O1 = 1st measure of the DV
X = IV
O2 = 2nd measure of the DV

Since both groups (for control of intervening variable) have their samples selected from a given population & then randomly assigned (R), they will be equivalent to each other & representative of the total population from which they were drawn. Pretests (O1) are given to both to ensure further, in case of sampling error, equivalence between the two groups. The posttest (O2) will determine whether the treatment or program (X) was effective if there is a significant change compared to the pretest in the experimental group. This time, there is a sound basis for the inference because intervening variables, such as in crease in the price of tobacco, would have affected both groups equally.

3.b. Longitudinal Control Group Posttest-Only Design

FORMULA
EG: R X O1
CG: R O1
Where: R = Random selection & assignment to group
X = IV
O1 = 1st & only measure of the DV

Two groups are used to control for intervening variables, but unlike 3.a above, no pretests are given because we simply assume (though, less certain than if pretests were performed) both groups are equivalent because they were produced by random sampling & random assignment (R) procedures.

Conclusion:
“None of the above designs are inferior to one another. Each has advantages in terms of time, cost, & data obtained” (p. 171).