EXPERIMENT DESIGN: Stage 3: Add All Conditions, Set Number of Trials, and Sampling [37330]

The goal of Stage 3 is to fully design the set of stimuli to be presented to the participant. This involves identifying the complete stimulus set and the manner in which the stimuli are sampled (sampling method). For this particular example, expanding the stimulus set also involves identifying a new independent variable. The four steps for Stage 3 are identified in the table below:

Stage 3, Step 1: Add All Conditions

In Stage 2, we added a Prime Display to the trial procedure. As is typical in priming studies, a priming stimulus is presented to the participant prior to the presentation of the probe stimulus. For the purposes of this example, we are keeping the Prime manipulation simple: the Prime stimulus is either the string “word” or “nonword”. More advanced priming versions of the lexical decision task might manipulate the nature of the Prime more extensively, for example by presenting 3 different prime types: words related to the probe, words unrelated to the probe, and nonword strings. Nevertheless, the Prime manipulation implemented in Stages 2 and 3 here illustrates the basics of how to include a priming display in the trial procedure, along with how to log the relevant independent variable information.

The new priming manipulation requires a change in the terminology with which we describe the trial stimuli. Back in Stage 1, we identified a single independent variable, Condition, which was used to identify the sole condition of the experiment: whether the Probe Display is a word or nonword. Now that we no longer have a single manipulation, using a single independent variable to define a trial no longer makes sense.

The new design is a 2 by 2 design which results in four types of trials. The first factor is the Prime type (the string “word” or “nonword”), and the second factor is the probe type (a word or nonword). The independent variable that we identified as “Condition” in Stage 1 now identifies the probe type, and we need a new independent variable that identifies the Prime type. (In this simplified design, the Prime type also happens to be the actual string that is presented to the participant on the Prime Display.) The combination of four trial types is shown below.

After identifying all of the PrimeTypes and ProbeTypes, we need to identify all possible independent variables. In other words, all possible combinations of the PrimeType and ProbeType that can constitute the independent variables. We will use two words and two nonwords to illustrate:

When you have a small number of independent variables/stimulus combinations they can be listed easily in a table as shown above. If instead there are a large number of independent variables/stimulus combinations, you may not want to create such a table. For example, if your study consisted of 50 unique word strings and 50 unique nonword strings, a table of all possible combinations of prime and probe would contain 200 rows!

Instead, if you are fully crossing all combinations of independent variables and stimuli, you may choose to identify the unique stimuli and then describe the method in which the stimuli are to be sampled and paired with the independent variables. Such a description might look like the following:

PrimeType - The list of primes to be presented is:

• Word
• NonWord

Stimulus - The list of stimuli to be presented is:

• cat
• dog
• jop
• fuame

Sampling – Each ProbeType is paired once with each Stimulus.

NOTE: Please view GETTING STARTED: Factor Table Wizard [37343] for information on how the Factor Table Wizard program included with E-Prime can help construct a full crossing of the factors and levels.

Stage 3, Step 2: Set the Weights
In some experiments, you may want to explicitly set the relative frequency of trial types. In the table of conditions that was shown above there were an equal number of word and non-word trials. However, we are going to change that specification, and present each of the non-word exemplars twice; this will result in the participant seeing twice as many non-word trials (8) as word trials (4). 

To change the relative frequency of trial types, change the weight of the rows. In this example, this is done by setting the values for the Weight attribute in the your List. Set the weights for the rows of ProbeType = “Word” to “1,” and the weights for the ProbeType NonWord rows to “2,” making the ratio of non-word to word stimuli 2 to 1.

Stage 3, Step 3: Set the Sampling Mode and Exit Condition
The sampling mode allows the altering of the order in which levels or conditions are run. Sampling modes include sequential, random (without replacement), random (with replacement), counterbalance, offset, and permutation. Each of these modes is described below.

 
The simplest sampling mode is sequential presentation of items. When sampling from your stimulus list this way, all participants see all of the stimuli in the same order as you enter them. While sampling a list of stimuli in sequential order can be very useful when debugging and testing an experiment, it is rarely used when collecting experimental data.

Random sampling, without replacement, is a frequently used sampling technique, and is what we will use for our lexical decision experiment. Stimuli are sampled randomly without replacement until the specified number of stimuli have been presented. For our example, we have 8 nonword and 4 word stimuli, for a total of 12 trials. After the last stimulus has been selected from the list, then all stimuli are re-randomized and become available for sampling again. If you set the number of trials equal to the total number of exemplars, then sampling randomly without replacement ensures that all participants see all exemplars within a single run of the experiment, and all participants see a different randomized order. For our lexical decision experiment, we will present a total of 24 trials, so the full list of exemplars will be sampled randomly without replacement until the 12 exemplars are sampled; then the exemplars will be returned to the pool of available samples, and will be sampled again randomly without replacement for another 12 trials.

Stimuli can also be sampled randomly, but with replacement. In such scenarios, each stimulus has an equal chance of being selected on a given trial. However, participants are not assured of seeing all exemplars, because after each trial the chosen exemplar is returned to the pool of eligible exemplars.

Another useful sampling technique is counterbalancing. When you counterbalance conditions, one exemplar is picked based on a selector variable, such as subject number or session number. For example, a design running six conditions with counterbalance by participant would result in only the third condition being presented to subject number 3. This might be used for a Latin Square design between participants.

In addition to counterbalancing, there are two other sampling methods that utilize a selector variable. When sampling with the offset method, the first exemplar sampled from the list of stimuli is determined by the offset factor, and then subsequent exemplars are sampled in fixed order from that location for the specified number of samples, and wrapping around to the beginning of the list when the end of the list is reached. For example, a design running six blocks of trials with offset by participant would result in the first sampled item being determined by the subject number. For subject number 3, the third item would be selected first; the list selection would then continue in fixed order (i.e., the fourth item would be selected next, followed by the fifth, etc.) until the end of the list is reached. The sampling would then wrap to continue with the first item in the list, then the second item, and conclude (because all of the items in the list have been sampled). This might be used for a Latin Square design within participant assignment of conditions.

Lastly, with the permutation sampling method, all possible combinations of conditions are created. Then, from the pool of possible combinations, one combination is chosen based on the value of the selector variable. For example, a design running three blocks of trials (A, B, and C) with Permutation by Participant would result in the generation of six possible combinations of conditions (i.e., ABC, ACB, BCA, BAC, CAB, CBA). From those possible combinations, subject number 3 would receive the third combination (i.e., BCA). Care should be taken when sampling with permutations, since the generation of all possible conditions increases factorially. That is, a large number of conditions will result in a tremendously large number of combinations of those conditions (e.g., 5 conditions result in 120 combinations, 6 conditions result in 720 combinations, etc.). The permutation option is best used with a small number of conditions.

NOTE: Interactive order is new to E-Prime 3.0 and may be selected as well (E-STUDIO: Interactive List [23589]). This order is specifically useful for debugging and MRI studies.

The table below illustrates how a List with three elements (values A, B, and C) might be sampled using the different methods described above. The last three methods (counterbalance, offset, and permutation) assume that subject number is used as the selector variable. For example, considering the counterbalance method, participants 1, 4, and 7 would sample from row 1, participants 2, 5, and 8 would sample from row 2, and participants 3, 6, and 9 would sample from row three. In contrast, the permutation condition with 3 conditions has 6 possible sequences. Participants 1, 7, and 13 would have the first order, participants 2, 8, and 14 would have the second, etc.

Stage 3, Step 4: Verify

As always, at the end of an incremental design stage, you should be able to envision the displays that the participant would see and anticipate how to analyze the data. The trial procedure was not altered in Stage 3. However, we added more stimuli and changed our sampling, so that a total of 24 trials are presented, in random order with non-words occurring twice as frequently as words.

The data analysis has changed, with the changes in the independent variables (renaming “Condition” to the more useful “ProbeType” and adding “PrimeType”). The table below shows the expected data file structure, with simulated values entered for the dependent variables:

Next Article: GETTING STARTED: Stage 4: Add Blocks and Block Conditions [37331]
Previous Article: EXPERIMENT DESIGN: Stage 2: Elaborate the Trial Procedure [37329]