Design of Experimental and Correlational Research Methods
One of the main aims of neuroimaging studies is to quantify relative changes in neural activity. In this respect, the contrast represents an attempt to identify the relationship between brain activity and the specific behavior or task the subject engaged in. If you think of the experiment as a matter of inputs and outputs, then the inputs would be the stimuli presented, tasks performed, instructions given, etc. The output would be the change (brain activation) that is then evoked in the brain. Thus, when those inputs vary, for example when an experimenter uses different modalities in the stimulus/response or the studies’ have different participant groups, these are important pieces of information that we need to document.
When entering data for the Contrast tab, questions about experimental design come into play- the more hypotheses being tested by the experimenter means more complicated designs (e.g., conjunction, factorial designs). Thus, it’s good to ask ourselves what the believed source of activation- Is it task-specific? Is it stimulus-driven? Careful consideration of design can also help us gauge what we need to document here- What did the experimenter seek to manipulate in choosing a paradigm (i.e., what varied the type of stimulus or the presentation of stimuli)? What was the condition of interest? Lastly, the analysis an experimenter conducted also allows us to glean group differences- Did the experimenter compute between group variability or within group variability? As you can quickly see, as the factors an experimenter examines increases, the number of components an observed activation can be attributed to grows as well.
Consequently, when entering data for the Contrast tab, questions about experimental design come into play. We’ll not enter into an in-depth explanation here of experimental design. Nor will you find detailed information here about the limitations of each type and we won’t pose one research design against another to illustrate the benefits of the former vs. the latter. What will follow is a brief overview of some well-utilized neuroimaging methods to give some background to the aforementioned point about design. The contrast is the heart of the experiment; it is the proposed relationship between cognitive architecture and neuronal activity and is directly related to an experimenter’s hypotheses. Ergo, to get the whole picture it’s important that we 1) highlight each and every relevant facet of the design and 2) try to account for all potential bias.
Move up from end of page: Hopefully, the preceding paragraphs have increased your understanding of what pieces of information are important in gleaning relevant metadata from each experiment and have driven home our initial assertion that the more complicated the hypothesized relationship between cognitive architecture and its neuronal implementation is, the harder our job as potential coders becomes in capturing all the important and sometimes complex details.
Experimental research designs manipulate a factor of interest. Overall experimental designs are generally one of three types: categorical, factorial or parametric.
Subtraction (A-B): This is one of the oldest and most basic experimental designs used to make inferences about cognitive processes; it's comprised of a cognitive component of interest and a baseline task (the task pair). It involves the statistical comparison of activations from an experimental condition with activations from a control condition. (T-tests are often used to analyze and detect differences and allow us to look at two levels of one independent variable (IV)).
Conjunction ((A-B) + (A-B)): Conjunction designs elaborate upon subtraction and are best thought of as the adding together several individual subtractions or task pairs (activation - baseline), to look for commonalities in activation patterns. Key to the conjunction method is a singular component or feature, that is isolated in each subtraction, i.e., revealed as being the difference in each A-B pair. In essence then, any observed activation involves the testing of many hypotheses conjointly; each having a significant effect that is additive and leads up to this collective activation. Conjunctions can be conducted across different types of tasks, stimuli, modalities (vision, audition), etc.
Factorial (A X B)
Factorial designs allow experimenters to look at several variables when explaining brain activation (e.g., average activity across different levels of a particular factor). In factorial designs, analyses of variance [ANOVA] are often utilized to detect differences e.g., one-way ANOVAs look at multiple levels of one IV, two-way ANOVAs look at multiple levels of two IV, etc.). Factorial designs have main effects and interaction effects. A main effect is the effect of an IV on a dependent variable (DV), ignoring the effects of other IV’s. Interaction effects are when two IVs interact and it is present when the impact of one factor depends on the level of the other factor. Factorial designs are powerful because they allow the experimenter to examine two or more IVs at the same time, while also assessing any interdependence between them.
Parametric designs vary an experimental parameter continuously and relate brain activation to this parameter. The experimental parameter can represent stimulus and task properties (e.g., lumosity, level of difficulty), stimulus presentation differences or performance (e.g., reaction time, accuracy). Parametric designs are good at identifying common areas of activation across different levels.
As opposed to experimental research where some factor of interest is manipulated, correlational research examines the effects of existing variations in a factor.
This type of design seeks to look at some behavior performed by the subject(s) and compares it to neuronal activity seen in a particular area of the brain; it does not explain any causal link between the two, it just looks at the relationship between them. Functional connectivity or the co-activation seen in different brain areas is also based on a correlational design, as is structural equation modeling which uses correlation data to study connectivity networks.
How a stimulus is presented to the subject/participant, is also an important piece of data that we document here in the Contrast tab:
Blocked (Box-car design)
Different conditions within an experiment are presented in separate blocks of trials (also known as an “AB Block”). Thus, the activation is seen across whatever interval of time the experimenter has chosen because each block is essentially one scan representing one condition. The blocks wherein some event (i.e., presentation of some stimulus) occurred invoking brain activation are then compared with baseline or control scans where it did not.
Event-related designs model signal changes associated with individual trials, as opposed to blocks of trials. Trials in which the condition of interest was not presented (or was off) are subtracted from the trials in which it was (on). This makes it possible to ascribe activation changes to particular events, permits randomization of stimuli, makes possible the retrospective assignment of trials (e.g., correct and incorrect responses), and generally just allows the experimenter to more closely examine the relationship between behavior and neural response.