Bananas in pyjamas?

Research News > September 2010
Statistics: Bananas in pyjamas? No, not B1 and B2, rather L1 and
In the July edition of Research News, Ingrid Burrowes wrote of ‘our complex language landscape' and
espoused the view that to ‘truly engage with Australia's population necessitates speaking a multitude
of languages' and further that ‘market researchers must take heed of (the associated) trends'.
In attending the Informs Marketing Science Conference in Köln, Germany, last June, amongst some
600 (!) individual presentations, I came across one that dealt with what would seem to be an
important aspect of this topic.
Stefano Puntoni and his colleagues at the Rotterdam School of Management, Erasmus University,
reflect on the fact that relative to a few decades ago, a much larger share of marketing research data
are now collected from ‘multilingual or multicultural' respondents. And although marketing research
agencies often translate surveys into respondents' native (L1) language, there are many instances in
which data are collected in a respondent's second (L2) language (typically English).
Hardly a path-breaking finding! However, what the authors have also established, through the conduct
of nine separate investigative studies covering more than 1,000 respondents, is a strong and
systematic tendency for individuals to report more intense emotions when answering questions using
L2 rating scales than when using L1 rating scales.
That is, L2 rating scales yield more extreme responses than L1 rating scales in the case of emotionladen items.
This phenomenon is known as the ‘Anchor Contraction Effect' (ACE). A couple of examples taken
from the paper serve to illustrate:
 offers customers the opportunity to rate any product using the emotional
statements ‘I hate it' and ‘I love it'. Regardless of their native language, people around the
world contribute ratings to the website and the status of the customer (i.e. whether L1 or L2)
may exert a significant influence on such ratings.
Hotels and other sites visited by international travellers typically field self-completion customer
satisfaction questionnaires. The emotional anchor point ‘Happy' (or similar) is often employed
in the response scales used. In such cases, ACE could lead foreign (L2) visitors to express
more positive opinions than local (L1) residents.
How should we correct for this introduced bias? Well, one approach (as one well-known and
respected academic said to me was his practice) might be to simply exclude from the data any
respondent who was too overtly L2, which (depending on the context) some might justifiably view as
being a somewhat extreme remedy.
Another obvious approach, offered by Puntoni et al, would be to make sure all respondents answer
items in their native language, although the authors acknowledge that to provide L1 rating scales to
everyone could be too costly or impractical.
In fact, the authors claim that ACE can be accounted for a priori with ‘corrective techniques'. Two
approaches that they claim to be ‘simple' are based on the ‘concomitant presentation of verbal and
non-verbal cues':
Emoticons can be used when measuring basic emotions that can easily be portrayed with
stylised facial expressions, and are particularly appropriate in online interview settings, and
also with children, those with low L2 proficiency, and those with low literacy.
Colours are claimed to be especially suitable in the case of abstract or complex emotional
concepts (e.g. by placing visual cues such as colour dots) of increasing intensity under the
response points along with the verbal labels), although they may be vulnerable to crosscultural differences in interpretation.
If neither of these approaches is deemed suitable (or if you think they are rubbish), the authors
suggest that researchers can adopt an a posteriori approach and use information about respondents'
L1 as a control variable (e.g. adding a dummy variable in regression models), but even this has
Whether or not you agree with any of the above (and I would urge you to read the full paper*
yourselves), it is clear that a respondent's L1/L2 status will impact on how (s)he responds. It would
seem to behove us as researchers to be cognisant of the ACE impact and either control it at the point
of data collection, or take it into account in our analysis (e.g. via separate analyses for the L1 and L2
subgroups in the sample).
Scott MacLean, Nulink Analytics
Footnote: Whilst it easy to say that academics can have a somewhat simplified view of the real world,
at least they do have the time to look at issues that we practitioners sometimes feel inclined or
compelled to gloss over. And whilst academic research can be a little unrealistic - e.g. to the point of
basing findings on convenience samples of at most a few hundred paid students - it often raises
issues that subsequently hit the research mainstream and help us in what we do in our day to day
* The full text of the paper can be found here - free to download but not to print: Publications/AMA Journals/Journal of
Marketing Research/JMRForthcomingArticles.aspx