Conjoint Recognition is a dual process model of recognition memory similar in flavor to process dissociation, but quite different in terms of its underlying conceptual framework. It's based on many of the ideas originally developed in Fuzzy Trace Theory.

In this summary I will follow the article in that I will (1) Describe Process Dissociation and the problems the authors have with that model (2) Describe Conjoint Recognition (3) Describe the author's experiments and (4) Describe what they say in the General Discussion about what you should take away from the article.
 

Brainerd et. al discuss Mandler's conception of recollection and familiarity.  Recollection according to Mandler involves interitem associations (associations between items) and familiarity involves intraitem associations (associations between the parts of individual items).  Arguably therefore, recollection should be influenced by manipulations that emphasize conceptual analysis and familiarity by those that emphasize perceptual analysis.

Process dissociation theory is a way of making sense of recognition memory data and separating conscious (recollection) from unconscious (familiarity) influences.

The theory assumes that:

• For any particular item on any test (whether direct or indirect), you can respond based on recollection or familiarity or a combination of the two.
• Recollection and familiarity are statistically independent.
• Recollection allows one to identify source information and thus select or reject an item based on what source it comes from. Familiarity does not allow one to identify source specific information.

The idea behind this is simple, items that are familiar should be selected regardless of the instructions, because familiarity does not allow one to distinguish between sources.  However, for items that are recollected, whether you select them or not depends on what instructions you've been given.

Example: Say 50 items are read to subjects, half by a male speaker and half by a female speaker. In the exclusion instructions you might be told, only select those words that were spoken by the female and DO NOT select any words spoken by the male. In the inclusion instructions you might be told, don't worry about who said the words, just select any word that was spoken by either speaker.

You are left with two important empirical probabilities. The probability of selecting words spoken by the male speaker when given the inclusion instructions (I). The probability of selecting words spoken by the male speaker when given the exclusion instructions (E). So the question is how can one figure out the parameters of having a recollection (R) and of something feeling familiar (F) from the empirical probabilities I and E.

Start by drawing a tree diagram of the possibilities
 

One thing that's nice about Process Dissociation is that you can figure out the parameters of the model just based on the empirical probabilities.  All you have to do is use a little algebra to solve simultaneous equations.  In fact it should be obvious to you that R=I-E (subtract the top equation from the bottom equation).  It should also be clear that F=E/(1-R).

Complaints About Process Dissociation

• The model has as many parameters (2: R and F) as there are empirical probabilities (2: E and I).  That makes it hard to falsify through traditional goodness of fit tests.
• Some researchers have argued that R and F are not always statistically independent.
• Some people have argued that R and F are unlikely to remain the same in the two instructional conditions.  For instance, you could imagine if people are given the inclusion instructions they might search memory less carefully, increasing the value of F and decreasing the value of R.
• Process dissociation does not include a measure of response bias (although recent extensions of the model do take response bias into account).

• In general, Brainerd et. al point out that recollection and familiarity cannot simply be a function of conceptual vs. perceptual processing.  Indeed, recollection is clearly influenced by manipulations of perceptual processes and familiarity can be influenced by conceptual processes.
• Argue that false memories cannot be accounted for through the recollection/familiarity distinction, at least not as proposed by Mandler because:
• Recollection should not support false memories but rather the rejection of non-presented items.
• Familiarity should not support false memories because familiarity is based on surface overlap not semantic overlap.

• Verbatim Representations: Verbatim representations represent information at the item level including surface features of the item.  If you see the word "DOG" your verbatim representation would be "DOG".
• Identity Judgments: An identity judgment is one in which the item on the test exactly matches your representation in memory.  You use identity as your decision criterion if you retrieve a verbatim trace.  If you retrieve the trace "DOG" the test item has to be exactly "DOG".
• Non identity Judgment: A nonidentity judgment occurs when you retrieve a verbatim trace that mismatches the test item and causes you to realize that the test item was not an item that has been previously presented.  Nonidentity is the process that produces false recognition reversal.  Say the test item is "CAT" and you realize, "No, it wasn't 'cat' that was presented, it was 'dog'".

• Gist Representation:  Gist representations represent information at the level of general senses and meanings.  If you see the word "DOG" your gist representation might be "I saw some sort of animal."
• Similarity Judgment: A similarity judgment is one in which you select the item on the test if it is similar to your representation in memory.  You use similarity as your decision criterion if you retrieve a gist trace.  If you retrieve the trace "some kind of animal" you will select "DOG" to the extent that it matches your prototype for the category animal.  Conjoint Recognition includes two similarity parameters.  One for the probability of making a similarity judgment to a target and one for the probability of making a similarity judgment to a related distractor.

• Response Bias:  Conjoint Recognition includes three response bias parameters.  These refer simply to the probability of selecting an item based on guessing, or some strategy like response alteration (I haven't picked one in a while, I don't really remember this one, but I better pick one soon, etc.)

• Targets: These are items that were presented to them (e.g. dog, chair, doctor)
• Related:  These are items that are semantically related to the presented items (e.g. cat, table, nurse)
• Unrelated: These are items that are not related to the presented items (e.g. hat, soda, pin)

• Target Instructions: These people are told to pick only targets
• Related Instructions: These people are told to pick only related items
• Target + Related: These people are told to pick both targets and related items

Click on the cell to see a description of the equation for each of the 9 empirical probabilities.
 

	TARGET INSTRUCTIONS	RELATED INSTRUCTIONS	TARGET+RELATED INSTRUCTIONS
Probability of picking a target	P(t|T)	P(t|R)	P(t|T+R)
Probability of picking a related	P(r|T)	P(r|R)	P(r|T+R)
Probability of picking an unrelated	P(u|T)	P(u|R)	P(u|T+R)

 

• Three experiments are reported.  In each experiment, subjects listened to a list of 80 familiar words.  Half the words were read once and half the words were read twice.
• After hearing the words subjects were given detailed instructions about their instructional condition (select Target, select Related Distractor, select both Targets and Related Distractors).
• Recognition tests included targets (40), related distractors(20) and unrelated distractors(20)
• Experiment 1: Related distractors were synonyms
• Experiment 2: Related distractors were antonymns
• Experiment 3: Related distractors were category names.  Experiment 3 also used the verbatim priming manipulation, where the related distractor was immediately preceded by the relevant target for half the items.

• Goodness of Fit Test: Measures the structural adequacy of the model.
• Goodness of fit tests are used to decide whether in comparing the model to the empirical data you are left with so much error that you can reject the null hypothesis that the empirical data could have been generated by the model?
• For all three experiments the goodness of fit tests were substantially below the level needed to reject the null hypothesis.
• They also performed monte carlo simulations to demonstrate the the goodness of fit values obtained pretty much matched up what you would expect based on the estimated parameter values and the number of subjects.
• Beyond global goodness of fit test, they compared all nine empirical probabilities individually to the values predicted based on the derived parameters and found a close match.
• Invariance Assumption:  If the parameters are invariant across instructional conditions, then it is possible to combine them in such a way that you can describe one empirical probability in terms of others.  If the parameters vary across instructional conditions the equalities will not hold up.  They were able to test 16 such equalities, and found that the differences were not more than one would expect by chance.
• Similarity/Identity Predictions(See Table 7)
• At short retention intervals, judgments should be primarily verbatim based. This prediction was confirmed in all three experiments.
• Repetition should increase verbatim memory more than gist memory (because verbatim fades more quickly).  Repetition increased the values of I and N by more than it increased the values of the similarity parameters.
• Antonyms are stronger associates than synonyms.  This should result in greater verbatim retrieval for antonyms than for synonyms, resulting in more non-identity judgments.  This prediction was confirmed.
• In the verbatim priming condition you should have more non-identity judgments.  This prediction was confirmed.
 

• Provides a model based framework for approaching dual process theories, that has advantages over process dissociation.
• Accounts for semantic false recognitions
• Places dual process theories within the identity / similarity rubric which they favor.