The elements and protocols that are included in the experiment in order to dissociate the contribution of the experimental variable from that of other factors.
In the Middle Ages 'control' meant a 'duplicate register', or 'duplicate roll' (contra-rotulus in Latin). Rolls were the standard material for writing administrative and financial records (Clanchy 1993), and the duplicate rolls were used by one officer of the law to check upon the rolls of another officer. The transition from the language of bureaucracy and commerce to that of scientific "culture has its roots in the Methods of experimental inquiry proposed by the British philosopher John Stuart Mill. One of these methods was 'The method of difference', whose canon reads as follows: 'If two or more instances in which the phenomenon occurs have only one circumstance in common, while two or more instances in which it does not occur have nothing in common save the absence of that circumstance, the circumstance in which alone the two sets of instances differ is the effect, or the cause, of an indispensable part of the cause, of the phenomenon' (Mill 1884). Hence, to demonstrate that a certain variable (the manipulated, or 'independent' variable) is the cause of an effect (on the 'dependent variable'), one should perform the experiment under two conditions, which should ideally be identical except for the omission of the change in the independent variable. The condition in which the change in the independent variable is omitted came to be known as the 'control' (Coover and Angell 1907). Controlled experiments may involve multiple treatments of the same "subject(s), 'experimental' vs. 'control', or different groups, ideally matched for their composition (e.g. gender, age, history), in which case we have the 'experimental group' vs. the 'control groups'. As it is usually impractical to have truly matched groups, the alternative is to assign subjects in a population to groups at random to smoothen variability (Fisher 1966). Statistics is then used to quantify the effect of the experimental variable, and further, to identify interactions among seemingly controlled variables, and to unveil potential latent variables (Fisher 1966; Martin and Bateson 1993; Kerlinger and Lee 2000; "dimension).
The use of controls in experimental science has started to gain popularity in early psychophysics (reviewed in Solomon 1949; Boring 1954; Dehue 1997), and later in educational research (Thorndike and Woodworth 1901 b; Coover and Angell 1907; Winch 1908). But the concept was occasionally appreciated by others as well: a "classic example is the gedanken experiment ('thought experiment') that proposed to test the effect of prayer on life in two groups of individuals, matched for everything except for practising 'prudent pious' (i.e. experimental) vs. 'prudent materialistic' (i.e. control) life-styles (Galton 1872).1 In the discipline of memory research, subjecting the same group alternately to experimental vs. control conditions is inherently problematic, as the mere experience may itself affect the outcome on subsequent tests. It is therefore advisable, whenever feasible, to use separate control groups. This, evidently, is easier said than done. Consider, for example, "functional neuroimaging of complex tasks in humans, or "delay tasks in "monkeys, in which suitable "subjects could be hard to get and train. Under such conditions, an experimental design with control groups often remains an Utopia, and protocols must be devised to reliably dissociate experimental and control responses in the same individual despite the expected order- and temporal experience-dependent effects (e.g. Buckner et al. 1996).
The canny use of controls is one of the most important tricks that newcomers to the experimental sciences must learn. On the one hand, the basic desire should be to control for whatever possible in order to make sure that the effect, if any, is indeed due to the independent variable, and to avoid embarrassing remarks of anonymous referees when the paper finally gets reviewed; on the other hand, multiple controls imply excessive expenses and delay in the completion of the work. Controlling the economy of controlled experimental designs is hence an art. In this context, 'internal controls' are a blessing: those are elements embedded in the protocol of the experiment that spare the need for separate controls. For example, suppose that we set out to test the effect of a circumscribed brain lesion on long-term memory. We could include in the protocol a test for short-term memory as well, because if long-term memory is impaired but short-term memory is not, we have an internal control for the lack of the effect of the lesion on the sensory and motor faculties that are required to perform the task. This eliminates the need for a separate control group to test the effect of the lesion on these faculties. A protocol well-designed may include a matrix of 'balanced' or 'reciprocal' controls, i.e. a combination of control groups or treatments that complement each other. A classic example is provided by the 'double-dissociation' lesion protocol (Teuber 1955). In this protocol the effect of two different circumscribed lesions, A and B, is tested on two different phenotypes, X and Y. If lesion A yields a defect in X but not in Y, while lesion B yields a defect in Y but not in X, then the defects in X and Y are not due to general damage but rather to specific dissociable contributions of A and B (see "declarative memory).
Memory research is of course not unique in benefiting from the proper application of controls in experimental design. Some elementary issues that require appropriate controls are shared by many other experimental disciplines. These issues include, for example, the need to distinguish correlation from causality ("criteria, "method). But several potential pitfalls do call for controls that to some degree or another are specific to the discipline of memory research. Here is a short list:
1. Controls for species-specific behaviour vs. learning. Motor patterns that emerge in training and testing may comprise an innate ("a priori) response to specific "stimuli ('sign stimuli') rather than learning (Breland and Breland 1961; Moore and Stuttard 1979; Wolfer et al. 1998). In parentheses, it is worthwhile to repeat here the ever-valid advice to those investigators who mingle with the behaviour of experimental subjects (Homo included): know your subject!
2. Controls for "development vs. learning. This is especially important in behaviours whose "capacity matures gradually, or materializes in restricted sensitive periods ("birdsong, "imprinting). The rule of thumb, however, is that a definitive dissociation of learning from development could be unrealistic, as both are interwoven in the emergence and refinement of even the most elementary sensory and motor faculties (e.g. Crair et al. 1998).
3. Controls for one type of learning vs. another. For example, is an alteration in behaviour that is obtained in a conditioning protocol due to the "association among stimuli, or to nonassociative processes ("sensitization, 'pseudoconditioning' in "classical conditioning; Rescorla 1967)? Is a particular conditioned response due to stimulus-stimulus ("classical) or stimulus-response ("instrumental) conditioning (Jenkins and Moore 1973)? And when a horse 'reads', is it because it has mastered human language, or because it has realized that responding to the subtle bodily gestures of its master yields a sweet reward ("Clever Hans; Pfungst 1911)?
4. Controls to distinguish learning from alterations in "performance due to changes in the experimental conditions. An often neglected need is to control for "state-dependent memory in studies that investigate the effect of drugs or of modulation of gene expression on memory. In such cases, if the subject flunks on the memory test, it could be a consequence of the change in its endogenous state, due to the presence of the specific drug or gene product in training but not in testing, rather than to memory failure (Overton 1964; also "context).
5. Controls for premature conclusions about the fate of memories. When "forgetting is detected, is it because the "engram has indeed vanished? A control test performed after a while may unveil 'spontaneous recovery'.
6. Other controls to unearth factors that could alter the behaviour but have nothing to do with learning and memory. For example, drug effects that result from the mere expectation by the subject that something will happen (a 'placebo effect', Swartzman and Burkell 1998; Kvavilashvili and Ellis 1999). Or lesion effects that are actually a consequence of the surgical procedures, not of damage to the targeted brain area. This is why lesion experiments always require 'sham' treatments, in which the skull is treated the same way as in the lesioned animal but no brain lesion is induced.
Occasionally what is expected to be a routine, boring control, "surprises and becomes a mind twister. Here is such a case. A subject is conditioned by a certain "reinforcement. A naive expectation is that the more intensive the reinforcement schedule, the stronger the memory. A simple set of controls would include a non-reinforced group, and, possibly, a group reinforced only occasionally. Alas, the group whose behaviour is reinforced only intermittently remembers the best (Humphreys 1939). This is the kind of controls that opens a whole new field of research (see "experimental extinction). Similarly, devising innovative controls has unveiled much about the sophistication of "classical conditioning (e.g. Kremer 1971). The take home message: don't look for excuses, add all the control groups you can think of, they may prove even more interesting than the experimental ones.
Selected associations: Artefact, Bias, Culture, Method
'This gedanken experiment finally made it, 129years later, into the 'functional neuroimaging lab (Azari et al. 2001).
Was this article helpful?