A Further Exploration of the Effects of Restoration Postscripts on Reactance

Abstract

This study examined the effectiveness of restoration of freedom postscripts at mitigating psychological reactance following recycling messages. Results of a 2 (freedom threat: high, low) × 2 (postscript: restoration, filler) plus 1 (offset no-message control) independent group experiment (N = 134) replicated prior findings demonstrating how freedom threats can increase reactance and subsequently reduce perceptions of message quality while increasing anger toward the message source. Furthermore, as predicted, following high-threat messages, a restoration postscript increased intentions to recycle, whereas following low-threat messages, no such differences were found.

Keywords:

• Freedom Threat
• Persuasion
• Reactance
• Restoration

Notes

1. Quick et al. (2015) used a different type of a restoration postscript, and the effects of their restoration approach produced null results; thus, we are not reviewing their findings further.

2. Magnitude scales have long been established as an effective instrument for the measurement of a wide range of phenomena, including loudness, weight, discomfort, pressure, and taste, among other variables (see Stevens, 1975). With ordinal measures, typically including 5- or 7-point Likert-type scales, a considerable amount of information is lost due to the limited response options. With magnitude scales, the response categories are unfixed, allowing for the true range of responses. Moreover, magnitude scales are better suited for the analyses requiring interval level data. (Note, regardless of how researchers treat them as such, most Likert-type scales are not true interval level measures.) For a more in-depth discussion of magnitude scales, see Lodge (1981).

3. The data were winsorized by recoding a variable’s scores to a lower value. An attempt was made to winsorize as little as possible. To ensure conservative winsorizing, the following steps were used. (1) The distribution of each variable was examined based on the frequencies of scores and the histogram. (2) If outliers were present, percentile values associated with the 95th, the 90th, the 85th, and the 80th percentile were generated. (3) Winsorizing the scores to the highest percentile was considered first. The majority of variables were winsorized to the 95th or 90th percentile of the original value.

4. For example, the variable measuring how irritated participants were after reading the message had original skewness of 5.83 with standard error of skewness of 0.21, which indicated a significant departure from normality. After transformation, the skewness became −0.26, indicating a substantial improvement. All transformation formulas are provided in the instrumentation section. Different transformation formulas were used for different variables to accommodate the specific type of skewness present in a given variable (for details on data transformations, see Fink, 2009).

5. Note that transformations affect the data in very predictable ways. In a transformation equation, Y* = (Y + k)^λ, where Y is the original variable, Y* is the transformed variable, and k is a constant, using λ < 1 is likely to result in a more symmetric distribution for positively skewed data, and using λ > 1 is likely to result in a more symmetric distribution for negatively skewed data (Fink, 2009). Importantly, although transformations affect the scale on which a variable is measured, they do not change “the relative difference between people for a given variable” (Field, 2013, p. 201).

6. Note that the differences between low-threat and high-threat condition reported above constitute a successful manipulation check as well as a demonstration of reactance effects. We further cross-validated these effects through the comparisons with the control condition. Significant contrast effects revealed that, relative to the control condition, high-threat messages also lead to higher perception of the threat to freedom (contrast estimate = 1.51, p < .001, control condition: M = −1.05, SD = 0.82, n = 13, vs. high threat: M = 0.56, SD = 0.83, n = 43), perceived anger (contrast estimate = 1.01, p = .001, control condition: M = −0.56, SD = 0.81, n = 13, vs. high threat: M = 0.41, SD = 1.00, n = 43), and negative cognitions (contrast estimate = 1.54, p < .001, control condition: M = 0.00, SD = 0.00, n = 13, vs. high threat: M = 1.56, SD = 1.50, n = 43).

7. Comparing the high-threat condition to the control condition revealed no significant differences in behavioral intention.