Research evidence going against the model, has come from Underwood. Underwood asked participants to detect digits presented on either the shadowed or the non-shadowed message. Naive participants detected on average 8% of the targets, but an experienced participant was able to detect 67%. This practice effect is difficult to fit within Broadbent’s model. That message is not only affected by physical characteristics, but also by relevance.
Support for Broadbent has come from; Broadbent’s split span experiments. Where participants were asked to recall digits presented repeatedly in pairs, which were asked to be recalled in two ways, ‘pair by pair’, or by ‘ ear by ear’. Broadbent found that ear-by-ear reports were easier for participants, and produced much more accurate responses than pair by pair. Which supports Broadbent, those channels cannot be switched.
Broadbent has been criticised, for its perception of channels. The belief of separate channels is easy to understand in the context of the dichotic listening task- where each represents a channel. However, this definition is not so clear in less artificial settings, such as listening to some singing on the radio or being able to pick out the words of the singer or the different instruments in the backing music.
Now to a late selection model, Deutsch’s late selection model. The model suggests, that both the unattended and attended message is sent to the sensory process store. Both messages are sent for full analysed for meaning. A filter is not required here, and both messages are sent for semantic analysis (the recognition process). Here the LTM store is connected, where relevant information is linked. Now the most important and relevant information determines the response. Selection here is dependant on salience (noticeable) and pertinence (relevance) according to the specifics of the person. According to Deutsch, even after a message is selected, channels can be switched. After messages are chosen at a later stage, the unattended message may not be remembered, as these messages are stored in the limited capacity processor (STM) which has a limited capacity of 7 2+/- bits. It is after the recognition process and between the response processes that the bottleneck occurs, a selection takes place.
Experimental evidence for this model has come from Lewis. Where Lewis presented single words to participants in a shadowing task. Some of the words on the unattended channels were semantically related to those being shadowed while others, which were completely unrelated. After, the experiment participants were unable to recall anything from the unattended ear, as the early selection model would have predicted. Yet, participants took longer to start shadowing a word on the attended channel if a semantically related word was present repeatedly in the other ear. Suggesting like Deutsch would, that analysis of the non-shadowed input slowed down the processing of the shadowed word.
Further evidence has come from Corteen and Wood and Corteen and Dunn. Participant’s given a shadowing task to complete. The findings of the study suggested that, the unconscious processing involved of the non-shadowed channel could involve semantic analysis. That both channels are fully given semantic analysis, which of course supports Deutsch’s model.
Evidence going against Deutsch’s model has come from Broadbent and Cherry’s studies. Broadbent’s split span experiments. Evidence from this experiment goes against Deutsch’s model, as it believes that channels cannot be switched which in the end Deutsch’s model believes. Further evidence going against Deutsch has come from Cherry’s binaural listening test. As it supports the idea of voice characteristics. As the finding’s showed that participants were not able to differentiate between the two voices. Which supports Broadbent, and obviously goes against Deutsch’s entire model.
A flaw in the study is that it has been found to have methodological difficulties. The theory has been proved to quite difficult to evaluate because of the problem of demonstrating unconscious processing. So further research cannot be done, towards the theory.
Support for this theory has come from, Parsimony and Tresiman evidence. Deutsch does not need a filter like Treisman’s model; Deutsch’s model looks at the simplistic explanation to describe, to what happens to messages. Research evidence has also come from Grey and Wallenberg, where selection takes place for meaning, and is not processed by channel by channel.