Chapter 11: Experimental Design

 

 

Study Questions

 

 

 

 

 

 

 

1.   Define an “experiment” in your own words.

An experiment is the designing and implementation of a manipulation of a variable which is performed in order to evaluate a hypothesis. Usually experiments are aimed at adding to generalizable knowledge.

 

 

 

 

 

 

2.   What are the differences between independent and dependent variables?

An independent variable is the aspect of the experiment that researchers manipulate. By changing the independent variable, the researcher presumes that the dependent variable (what is being measured) will also change.

The dependent variable in a study is the quantity that is measured in order to determine the effect of changing the independent variable.

 

 

 

 

 

 

3.   What is a research hypothesis? What three basic types of hypotheses

      are possible for fMRI studies, and what are their characteristics?

A hypothesis is a statement about the nature of the world that makes predictions about the results of an experiment. A scientific hypothesis must be able to be proven false, or else it is not amenable to scientific investigation.

 

In fMRI there are hypotheses about: 1) hemodynamic activity, which examine the BOLD effect itself, without addressing the question of causation; 2) neuronal activity, which is estimated by transforming the measured BOLD signal, as neuronal activity cannot be measured by fMRI directly. These types of hypotheses use the BOLD signal to infer neuronal activity. 3) psychological hypotheses, which rely upon broader concepts that can be difficult to define, and are therefore difficult to formulate, but are also influential in forming theories.

 

 

 

4.   Why do some people consider fMRI data to be epiphenomenal? What do you think about this issue?

An epiphenomenon means that something (in this case, BOLD activity and fMRI data) is a result of some side-effect or primary phenomenon, instead of resulting from the presumed phenomenon (in this case, neuronal activity). In other words, correlation does not entail causation, and some people think that BOLD activity and neuronal activity are not causally related. I think this is a criticism that deserves more attention, because there are areas to explore that could at least help to settle some of this debate. 

 

 

 

 

 

 

5.   What are confounding factors in an experiment, and how can they be minimized?

A confounding factor is a factor that changes along with the independent variable and makes attribution of changes in the dependent variable more complicated (because attributing change in the dependent variable to the change in the independent variable does not eliminate the possibility that the confounding factor actually caused the change). To minimize confounding factors, experimenters should ensure that randomization is complete (i.e. ensuring that factors that vary within the experiment change in a random way with respect to the independent variable). Experimenters can also use a technique called counterbalancing, to equalize the influence of confounding factors across conditions of the independent variable.

 

 

 

 

 

 

6.   What are the basic principles of blocked designs? Why are they sometimes referred to as “subtractive” designs? A blocked design compares an experimental condition that is assumed to be present at all times for a defined period of time to a control condition or a different experimental condition. Basically, an experiment is divided into blocks of time, and the conditions within each block alternate. The data can be easily analyzed by comparing the dependent measure in each block condition.  They are referred to as subtractive because conditions are compared to each other by analyzing which voxels are active in one condition that are not active in another condition. In other words, BOLD activity differences between the two conditions are determined by subtracting one condition from the other. 

 

 

 

 

 

 

7.   Are blocked designs better for detection of activity or estimation of the time course of activity? Why?

In general, they are better at detection of activity, because within a single block (a single experimental condition) the processes may actually be very different and change over time. However, a blocked design only compares the blocks to other blocks. Also, the shape of the hemodynamic response is not examined as closely in blocked designs as in event-related designs, which means that the order in which voxels showed activity cannot be determined from the data.

 

 

 

 

 

 

8.   Why do some experiments evoke increased activity during control conditions compared to experimental conditions? What might subjects be doing/thinking during control blocks? It is possible that in certain experiments, the experimental condition involves inhibition rather than activation of certain processes, so that less activity in some areas of the brain occurs during the experimental task. Some researchers have speculated that introspection during ‘rest’ conditions account for the increase in activity of some areas.

 

 

 

 

 

 

9.   What physiological measure did Gusnard and Raichle suggest as an index of baseline activity in the human brain? The oxygen extraction fraction, which is the proportion of oxygen extracted from the blood. This measure remains stable across brain regions, so that the proportion of oxygen extracted during a baseline condition is basically spatially uniform.

 

 

 

10. Which brain regions show increased activity during control conditions?

The lateral parietal cortex (angular gyrus), the posterior cingulated or precuneus, the superior frontal gyrus, and the ventral prefrontal cortex

 

 

 

 

 

 

11. What are the basic principles of event-related designs?

Event-related designs assume that neural responses to stimuli in the experiment will be brief and discrete.

Event-related designs measure responses in the brain over time, as opposed to steady state brain activity measured in blocked designs. The detection power and temporal resolution are the key factors in event-related design, while spatial resolution is less important. The hypothesized duration of the hemodynamic response is more important in this type of design than in others.

Each stimulus is considered an impulse, which generates a hemodynamic response. These responses are assumed to conform to the linear systems model, meaning that BOLD responses to successive stimuli do not interact with one another.

 

 

 

 

 

 

12. What does the term “epoch” describe in an event-related design? An epoch is a segment of an overall time series or set of images that is separated from the full series and time-locked to a particular event. It usually is a period surrounding the event of interest (before, during, and after).

 

 

 

 

 

 

13. How do researchers often improve signal-to-noise in event-related designs?

By adjusting the interstimulus intervals.

 

 

 

 

 

 

14. What is the advantage of a prestimulus baseline period for event-related analyses? This allows analysis of the differences between two experimental conditions, as well as comparisons of each condition with the baseline. Otherwise only the relative difference between the two experimental conditions could be analyzed, which may not provide much information about the process overall.

 

 

 

15. What is the primary disadvantage of slow event-related designs?

Their inefficiency, due to the low density of events over time (requiring longer scan).

 

 

 

 

 

 

16. What sorts of experimental questions can be answered by event-related designs, but not by blocked designs? In general, event-related designs have better estimation power, and is better able to characterize the timing of events in the brain (as opposed to spatial properties). Event-related designs are important for questions about the timing of the involvement of brain areas for a task, and are helpful in identifying cognitive processes associated with distinct time periods. They also allow trial sorting, unlike blocked designs.

 

 

 

17.  What is trial sorting? In what sorts of experiments would it be useful?

Trial sorting refers to the analysis of data in a different way than was originally planned for—in other words, after the experiment has been carried out, one can examine the same data by reassigning events to condition. This is useful in experiments that provide additional sources of data, for instance when behavioral data are collected along with imaging data. It is also used to analyze data based on the responses of subjects, such as error rates or response time.

 

 

 

18.  What are semirandom designs? What advantages do they provide?

A semirandom design is one that varies the probability of the events or stimuli that occur. It is a type of event-related design that uses fixed intervals, but within those intervals the events occur randomly, so that in one interval many stimuli may occur while in the next only one may occur. This design can optimize the detection and estimation power, which are at odds with each other, and should be used in studies that value both detection and estimation equally.

 

 

 

19.  What are mixed designs? Why would they be used?

A mixed design combines blocked and event-related designs into a single study, typically by having blocks within which multiple types of events occur instead of only one condition per block. A mixed design allows statistical analysis of independent variables that do not co-vary. Each block is associated with a broader shift in strategy (e.g. attention), and analysis that compares blocks can thus measure a state-related process. A mixed design is therefore chosen when a sustained change in brain activity is the target of the investigation.

Event-related designs, and the short-term changes they induce in the brain, contain individual events that can measure item-related processes, which are changes associated with a particular stimuli. When research questions involve both types of processes, studies can be designed to incorporate both state and item-related processes.