Research Proposal

PSY 550 Topic 7 Resource

Introduction

In this topic, different types of correlated groups and developmental designs are examined. Correlated groups are where the experimental and control group are linked in some way. The matched-participants design is one type. The type of correlated design discussed this week is within-participants designs. In addition, advanced experimental designs, including multiple-group and factorial designs, are discussed. Finally, descriptive statistics are examined.

Within-Participants Designs

Within-participants designs (also called within-subjects designs or repeated-measures designs) are where subjects are reused in two or more treatments. These designs have their own unique strengths and weaknesses.
Subjects can be used in studies with one independent variable that has two or more levels. Subjects may encounter each level. For example, if the independent variable has two levels, A and B, they would experience A then B. To counter for order effects, a better design would be A, then B, then B, then A or ABBA.
In within-subjects factorial designs, subjects participate in studies with two or more independent variables. For example, if each subject has to rate four photographs of men showing different emotions and four photographs of women showing different emotions, that would be a 4 X 2 design (# of emotions X # sexes); therefore, each subject participates in eight trials. Clearly, a risk with this type of design is that subjects will start getting fatigued or sloppy as the number of trials increases.
In mixed designs, between-subjects and within-subjects variables are assessed in one design. For example, sex of the subject (a between-subjects variable) can be examined on a task involving repeated measures (e.g., subjects smell four different fragrances); that would be a 2 X 4 mixed design.

Advantages and Disadvantages of Within-Participants Designs

In many ways, within-subjects designs function as the ultimate matched design, where subjects serve as their own controls. This strategy helps eliminate random error because who is more like a person than himself or herself? In other words, this strategy controls for subject variables (Myers & Hansen, 2006). Because of reduced error variance, this type of design increases the power of statistical interpretation to detect a genuine effect of the independent variable. This type of design is perfect for testing medical treatments; for example, at time one, the subject’s blood pressure or cholesterol level is measured. The subject begins taking medication, and at time two (perhaps weeks later), a second measure is taken to see if the blood pressure or cholesterol level has dropped. Another advantage is that this design requires fewer subjects to recruit and train (Martin, 1977). For example, think of the 4 X 2 design listed above; if that were a between-subjects design, then 8 cells at 20 subjects per cell would require 160 subjects. In a within-subjects design, each subject participates in all 8 trials, so only 20 subjects are needed.
The disadvantages are that subjects may become fatigued or bored with participation in multiple treatments. The opposite problem is a practice effect, where subjects do better on successive trials because of practice. Both of these types of error result in increasing error (called progressive error) as subjects proceed from one trial to the next. The order of the treatments may be another potential confound; if one treatment brings about interference, a change (for better or worse) may occur in the next treatment. These are called order effects. Suppose a soft drink company wants to pit their product, Cola X, against an established product, Cola Y. They recruit subjects in Phoenix during the summer, and have them stand outside in the heat. Then the subjects come inside and are asked to drink 12 ounces of Cola X. Do you think they will like it? What will their reaction be to Cola Y, which they have to drink 10 minutes later? Is there a problem with the order of the treatments? That is why counterbalancing is necessary in within-subjects designs.
Counterbalancing may occur in different ways. Subject-by-subject counterbalancing requires every subject to participate in every trial (e.g., the ABBA design mentioned above). A second method, across-subjects counterbalancing, randomizes progressive errors, either through complete counterbalancing or partial counterbalancing. With two treatment conditions, complete counterbalancing is accomplished through half of the subjects receiving AB, and the other half receiving BA. The formula for treatment sequences is n! (n factorial), where n = number of treatment conditions. Thus, when n = 3, n! = 3 X 2 X 1, or 6. With greater numbers of treatment conditions, complete counterbalancing becomes difficult, if not impossible (e.g., 5! = 120 treatment sequences). Thus, partial counterbalancing becomes the solution and can be either randomized or accomplished through the Latin-square method (Jackson, 2011); this method chooses four orders out of the 24 that are possible. One potential problem is if condition A affects condition B, or C affects D, they each appear twice in a Latin square, and thus may show a greater potential bias than what really exists. The balanced Latin square corrects for this potential problem (Myers & Hansen, 2006).
With this type of design, one should always be aware of the potential of carryover effects, where exposure to one treatment weakens or invalidates performance in the next one (Jackson, 2011). This type of problem should always be tested with the counterbalancing measures.

Developmental Designs

Developmental designs use age as an independent variable. Longitudinal designs are studies that are carried out over a long period. This design is best used for examining developmental differences across time. Probably the best known longitudinal study is Terman’s study of gifted children, which began in the 1920s and is set to conclude in the year 2020 (ERIC Clearinghouse, 1998). The advantage of this type of study is that the same subjects are followed across time; this makes it easier to track developmental changes. The disadvantages include: 1) keeping track of subjects more than once (subject mortality occurs when subjects drop out of a study), and 2) the length of time needed to complete one project (in Terman’s study, nearly 100 years!). In contrast, cross-sectional studies can examine several age groups at once, so the timeframe is shorter. Obviously, however, since the groups are different to begin with, it is hard to know if significant results are due to developmental differences or due to subject differences (Santrock, 1990). Another problem with comparing different age groups is the cohort effect. A cohort is a group of people born around the same time; therefore, they have many similarities that other generations might not have. The third developmental design, the sequential design, combines features of the other two designs; it minimizes cohort effects but is the most expensive and time-consuming of these designs (Jackson, 2011).

Descriptive

are quantitative measurements of samples (Myers & Hansen, 2006). Descriptive statistics summarize group data in different ways, whereas inferential statistics make inferences about the probability that the observed finding was caused by the experimental manipulation of the independent variable (Martin, 1977).
Data may be represented in graphs. Bar graphs are used for discrete variables; therefore, the bars in these graphs do not touch (see Jackson, 2011, Figure 15.1, p. 219). In contrast, the bars in graphs for continuous variables do touch and may be represented as histograms (Figure 15.2, p. 219) or as frequency polygons (Figure 15.3, p. 220).

Measures of central tendency summarize what is typical in a distribution of scores. A normal distribution (or bell-shaped curve) is a bisymmetrical distribution of scores, where the mean, median, and mode are the same number. If the distribution is not normally distributed, then it is skewed (Myers & Hansen, 2006). The mean is the arithmetic average (total summation of test scores divided by number of scores), the median is the halfway point of the distribution (or 50th percentile), and the mode is the most frequently occurring score.

Measures of variability depict how spread out the distribution is. For example, the range is obtained by subtracting the lowest score in a distribution from the highest score and adding 1 (Bluman, 1998). Variance and standard deviation are measures of how scores are dispersed around the mean. The standard deviation is the square root of variance (see Jackson, 2011, Figure 15.9, p. 228 for the formula). The numerator of the formula, Σ (X – M)2, is called the sum of squares.

Setting Up the Review and Method Sections

As students contemplate their research proposal project paper, they need to consider several ingredients for their project. The first section of a research proposal includes a literature review and a section for the hypothesis. A literature review consists of experimental and nonexperimental studies relevant to the topic. The student should point out his/her proposal not only follows from previous research but also leads into new territory. Thus, the last part of the literature review is where the student lists his/her hypotheses.
The second part of the proposal, titled the method section, should consist of several subsections. 1) Subjects (or Participants): This is where the student should discuss how subjects will be recruited, how many are needed, and whether specific criteria are needed (age, sex, etc.). 2) Apparatus and Materials/Instruments: What types of measurements are needed? Common items (such as rulers and paper) do not need to be mentioned, but more sophisticated instruments do. If a test is going to be used, it should be mentioned here, whether it is a published or a homemade test. If it is a published test, copyright permission should be obtained, and reliability and validity coefficients should be reported (Buros Center for Testing, 2004). 3) Procedure: In this section, step-by-step details are needed in chronological order; the text of any instructions given to subjects should be included. 4) Design: The type of design is not usually included for simple studies, but it should be included for a project like this.

Setting Up the Results and Discussion Sections in the Research Proposal

Since you will not typically be carrying out your research proposal, you may be wondering how you will fill out your results section. There are four items that can be included. 1) Since you created the design for the Method section, you can select the appropriate statistic. 2) You can include the alpha level (of .05). 3) You know how many subjects you need, so df can be calculated. 4) Thus, you can also calculate the critical value needed to reject the null hypothesis.
The Discussion section should explain three aspects: 1) what the results mean, in “everyday” language (practical versus research significance), 2) the strengths and the limitations of the study, and 3) future possibilities for research (Martin, 1977). In addition, for this class’ project, a discussion of ethics is needed (e.g., how ethical issues were dealt with).
Following the body of the paper is the References page; see the American Psychological Association (APA) Manual for the correct format (pp. 49-51). The last two items are Tables and Figures. The research project must contain one of each (or two tables or two figures). Tables are columns of data (if you do not carry out the project, do not make up data; the table format can still be set up). Figures include pictures, graphs, or drawings.

Conclusion

This topic discussed how to create within-subjects designs, developmental designs, and advanced experimental designs. It also introduced the student to statistics, a topic that will be discussed more thoroughly in the next module.

References

Bluman, A. G. (1998). Elementary statistics (3rd ed.). Boston, MA: WB/McGraw-Hill.
Buros Center for Testing. (2004). Retrieved from http://www.unl.edu/buros/
ERIC Clearinghouse on Disabilities and Gifted Education. (1998). Gifted-longitudinal studies FAQ. Retrieved from http://www.hoagiesgifted.org/eric/faq/gt-long.html
Jackson, S. L. (2011). Research methods: A modular approach (2nd ed.). Belmont, CA: Wadsworth/Cengage Learning.
Martin, D. W. (1977). Doing psychology experiments. Monterey, CA: Brooks/Cole.
Matheson, D. W., Bruce, R. L., & Beauchamp, K. L. (1974). Introduction to experimental psychology (2nd ed.). New York, NY: Holt, Rinehart, & Winston.
Myers, A., & Hansen, C. (2006). Experimental psychology (6th ed.). Belmont, CA: Thomson/Wadsworth Publishing Co.
Privitera, J. G. (2017). Research methods for the behavioral sciences (2nd ed.). Los Angeles, CA: Sage.
Santrock, J. W. (1990). Children (2nd ed.). Dubuque, IA: William C. Brown Publishers.

Copyright 2010. Grand Canyon University. All Rights Reserved.