Download presentation
Presentation is loading. Please wait.
Published byTimothy Parrish Modified over 6 years ago
1
CSD 5100 Introduction to Research Methods in CSD
Group Designs in CSD Research Experimental Design and a Little About Data Analysis The Class Research Project: Methods
2
Research Designs Plans that include protocols for Selecting subjects
Controlling extraneous variables Variables other than the independent variable(s) we’re interested in that may affect the dependent variable Observing the dependent variable Ensuring ethical procedures Remember from last week, the research strategy was the overall plan of attack, the research design are the specific strategies to address the research question…
3
Group Research Designs
Designed to determine the effect of one or more independent variables on a selected dependent variable by comparing performance, observations, whatever among groups of subjects General definition of a group design…Lots of permutations of this design…can range from very simple to very complex. Complexity directly stems from the # of independent variables involved AND the number of levels they each have… This will also tend to guide the subject selection methods of the groups. The major idea is that the groups must be homogenous (alike) in every way EXCEPT the level of independent variable they represent.. Let’s look at some examples…
4
Two Group Research Designs
Randomized controlled two group design These designs represent the most basic of group designs. The design is meant to address the effect of 2 levels of one independent variable on a dependent variable. An example, let’s say we want to measure differences in perfomance of some “treatment” on a group of subjects. One way to do this using a 2 group design uses the following (fig 4.3). Here we identify a group or sample of participants—previous research might guide the experimenter in identifying the characteristics of the participants. The first group really represents a sample of a population you are interested in studying. In this example, the first group in randomly put into 1 of 2 groups representing the 2 levels of the independent variable (treatment vs no treatment). The key to random assignment is that it must be random—flipping a coin, table of random #s For statistical purposes, then, the influence of any other nuisance or confounding varibles that might impact the effect of the treatment will be reduced because the level of that variable will vary randomly in each group…give an example using gender as a nuisance variable.
5
Two Group Research Designs
Quasi-experimental two-group design Fig 4.4 In this example, we still have 2 groups reflecting the 2 levels of a single independent variable, and we will still compare the outcomes from the 2 groups to decide if there’s an effect of the treatment variable on the dependent variable. However, to ensure group equivalence, except for exposure to treatment, we use matching rather than randomized assignment. In this procedure, the experimenter ids critical matching variables like, age, IQ, SES, whatever…and makes sure each member of the matched or control group is exactly like the target or treatment group. Then the comparison of performance is made. The strength of group equivalence is the key to these designs. If subjects are assigned randomly, then what we know about randomization will ensure group equivalence. In the case of quasi-exp designs, equivalence depends on the matching procedures you use.
6
Achieving Group Equivalence
Two steps are necessary Identify the relevant variables known to be related to the dependent variables Match the groups of participants on the basis of these variables
7
Complex Group Designs Complexity of the basic group design can be increased by Having more than two levels to one independent variable Having more than one independent variable All the above
8
Multivalent Research Designs
Fig 4.9 This is used when there’s still just 1 independent variable, but more than 2 levels. Here’s an ex… Researchers interested in the effect of wearing Has on other people’s perception of “achievement” Had volunteers look at a person (photograph) wearing a body HA, a BTE HA, and no HA—the independent variable here is HA condition….there are 3 levels. The volunteers had to rate each photo in terms of “achievement) Results…
9
Factorial Research Designs
Attempt to model real-life events or conditions in which multiple variables may affect behaviors jointly Combinations of two or more variables sometimes produces a change in behavior that’s not present when the effects of single independent variables are observed Designs used when there is more than one independent variable…and usually more than 1 level of each. This is a very common and very powerful research design used in CSD…if the study employs a group design This is the kind of design our class project will use…
10
Classification of Factorial Designs
Factorial research designs are generally classified in one of these ways Related designs Each subject experiences all treatment conditions Independent designs Matched subject groups experience a single treatment condition (1 level of 1 independent variable) This is what our class project design will be Mixed designs Combination of related and independent designs Related designs was an example of my dissertation I told you about last week. Remember, I had NHLs listening to nonsense syllables in 3 levels of reverberation, 4 levels of filtering, about 5 levels of background noise, and 3 presentation levels. Each subject (I had 7 of them) listened to all possible combinations of all conditions (3 x 4 x 5 x 3=180 conditions) Independent designs is what our class project will be
11
Factorial Design Model
Example of a 2 X 2 data summary table I usually refer this as the ANOVA summary table or data table. It’s a way of thinking about how your data will be arranged for future analysis. Factorial designs are almost always analyzed using analysis of variance procedures (ANOVA), and this is the way the ANOVA prefers to see its data… This is an example of a 2 X 2 factorial design…Two levels of Ind var A and two levels of ind var b. The boxes reflect the dependent variable values collected by the subjects all exposed to the same independent variables and levels. Depending on how many subjects you have for each box reflects the number of dependent variables within a single box.
12
Factorial Design Model for the Class Project
Gender Male Female Age Young Kids Vot 1 Vot 2 Vot 3 Vot n Older Adults This is a 3 X 2 independent factorial design Explain the set-up In naming the design, the number of “X” equals or refers to the # of independent variables. The lowest you can have is 2. So, 1 “X” is 2 ind variables…A X B X C would be three indep variables, and so on. The numbers reflects the # of levels of the independent variable…So in our class example, “3” reflects the independent variable of age and “2” the independent variable of gender. Now, we really have a third independent variable, classification of consonants in terms of voice/voiceless. Remember we are going to eventually be measuring the vot times of 6 stop consonants, 3 are voiced and 3 are unvoiced. Now, we could just throw all six in one data cell…but we know that vot for voiced stops will be very different than vot for voiceless stops, so having both in the same data cell might not be a good idea. We also saw from previous research, that the voice/voiceless variable interacted in a wierd way with gender and age, so another reason not to throw them all in the same data cell. Another idea is to add another independent variable to our factorial design
13
Factorial Design Model for the Class Project Adding a Third Independent Variable
Gender Male Female Age Young Kids Voiced vot1 Voiced vot2 Voiced votn Unvoiced vot1 Unvoiced vot2 Unvoiced votn Older Adults This would be a 3 X 2 x 2 independent factorial design Here’s the data table if we did this… We’re not doing this for a couple of reasons.. Having 3 independ variables in a factorial design makes it too complex for our purposes. The number of interactions tested by the ANOVA will increase, as well as the number of main effects. We want to keep the analysis fairly straight forward your first time through Adding a third indep variable will reduce the number of observations (subjects)/cell. This will reduce the ability of the analysis to find a significant difference between the cells because you just don’t have a sufficient number of observations We already know, from previous research and from our knowledge of speech science/speech acoustics that vot times between voiced and voiceless stops are different—significantly different. We don’t need to reinvent the wheel here, and frankly, this part of the analysis will not be interesting, nor does it directly address our research question. Remember that our research ? Is interested only in the effect of age and gender on vot, not age, gender, voicing on vot. So, how should we view the data summary??
14
Class Project Data Analysis Model
Voiced Consonants Gender Male Female Age Young Kids Vot 1 Vot 2 Vot 3 Vot n Older Adults Voiceless Consonants Gender Male Female Age Young Kids Vot 1 Vot 2 Vot 3 Vot n Older Adults To “get rid” of this uninteresting variable (voicing), we will be doing separate analyses…one for the voiced consonants, and one for the voiceless. This way we will reduce the variability of observations within each subject cell AND at least indirectly look at the effect of the voicing feature by looking to see if the interactions observed are the same or different across the 2 analyses…
15
Class Research Project: Methods
Subjects Stimuli Data collection methods Instrumentation Measurements of voice onset time Anything else? Comparative ex, syntactic perf of kids with and w/out artic disorder. Multiple dependent variables can be used.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.