More Complex than Previously Thought: New Insights into The Optimal Administration of the Initial Preference Task

Abstract

The Initial Preference Task (IPT) is based on the name-letter effect (NLE: individuals prefer name over non-name letters) and is widely used for measuring implicit (automatic) self-esteem. However, its implicitness has been criticized because up to 85% of individuals can become aware of its self-relevant nature (i.e., recognizers; Krizan, 2008). The present experiment assessed differences in name-letter awareness across three different IPT administrations as well as IPT scores across recognizers and non-recognizers. A more elaborated administration (incorporating symbols as filler items) significantly reduced name-letter awareness; furthermore, IPT scores differed between recognizers (higher scores) and non-recognizers (smaller scores). Implications and limitations are discussed.

Keywords:

• Administration
• Initial preference task
• Implicit self-esteem
• Name-letter effect
• Recognizers

Notes

In line with De Houwer et al. (2009), we use the terms direct and indirect to refer to measurement procedures and the terms explicit and implicit to refer to their measurement outcomes. For example, the Rosenberg Self-Esteem Scale (Rosenberg, 1965) is a direct measure of self-esteem because it directly asks participants about their self-esteem. The IPT, in contrast, is an indirect measure of self-esteem because participants are not asked directly to provide information about their self-esteem but rather indirectly to rate the liking of letters. However, it is important to note that indirect measurement does not necessarily capture only implicit (i.e., automatic, unconscious) aspects and direct measurement procedures do not necessarily assess only explicit (i.e., conscious, deliberative) aspects.

Indeed, Krizan (2008) found that participants recognizing their initials during the rating process (so-called recognizers) consistently showed substantially higher name-letter effects as compared to non-recognizers.

It should be stressed that in the present study implicit–explicit correlations are not employed to establish the validity of the different measures of implicit self-esteem (i.e., correlations alone would be hardly sufficient), but to establish whether experimental manipulations also lead to changes in levels of explicitness/implicitness of the measures.

Participants in the symbol group had to do more ratings than in the remaining two groups because they additionally had to rate symbols. For comparability reasons, we only analysed the letters for validity; we also used only the letters and not the symbols for the calculation of the respective IPT scores.

This improved algorithm can be requested from the authors on demand.

In contrast to our study, Krizan (2008) only identified users stating their initial letters as recognizers, whereas we looked at all name letters (including non-initial letters). Yet, it turned out that only seven participants stated letters matching their non-initial letters and not their initials. Therefore it can be concluded that almost all recognizers stated that their initial (and not their non-initial) name letters had a special meaning for them during the IPT.

Eta-squared around and lower than .01 are considered as small, around .06 as medium, and around or higher than .14 as large (Cohen, 1988).

Throughout the remainder of this article, we will use the term name-letter effect (NLE) for IPT scores that are significantly different from zero, representing the phenomenon that individuals are significantly preferring name letters to non-name letters.

It should be noted that Krizan (2008) used the Self-Liking Self-Competence scale (Tafarodi & Swann, 1995) for measuring explicit self-esteem whereas we used the RSES.

All analyses were also calculated with other published (but not recommended) algorithms (see LeBel & Gawronski, 2009, for a review of algorithms). Basically, the algorithm-specific results were homogeneous, except for some single significant differences compared to the used I algorithm, which might be due to inflated Type-I error rates because of the large number of significance tests. All in all, there was no single algorithm showing substantial differences to the I algorithm throughout all analyses.