Participants
Participants were 186 mothers (Mage = 33.17 years, SD = 4.83) with at least one child between the ages of 7 and 12 (Mchildage = 8.95 years; SD = 1.61; 55.4% male, 43.5% female, 1.1% gender not listed). The majority of participants identified as White (93.2%), Non-Hispanic/Latino (94.4%), and heterosexual/straight (73.4%). Participants were recruited using a research database maintained by the psychology department of a public university in the northwest United States, which contains contact information for local families who have consented to be contacted about opportunities to participate in developmental psychology research (see https://teamduckling.uoregon.edu/ for more information). Additional eligibility criteria to participate in the study included being at least 18 years of age, identifying as a woman, and being a primary caregiver of a child. Recruitment was stratified based on income, such that study participants fell within one of three household income brackets: 1) below $35,000 (24.7%), 2) $35,000-$75,000 (38.2%), and 3) above $75,000 (37.1%). The recruitment process for this study and sample demographics are described further in [17].
Procedure
Interested participants completed a screening survey to determine eligibility. Eligible participants were directed to the study survey, which was created using Qualtrics survey software. Participants completed the survey using an electronic device at their preferred time and location. The survey took approximately one hour to complete. Participants were compensated with a $15 Amazon gift card. Informed consent was obtained from all participants, and study procedures were approved by the affiliated university’s Institutional Review Board.
Measures
Maternal proclivity to apologize
Mothers’ proclivity to apologize to their child was measured using the Proclivity to Apologize Measure-Parent (PAM-P; [18]). The PAM-P was created by adapting the pre-existing, general Proclivity to Apologize measure (PAM; [13]). In the adaptation, Ruckstaetter and colleagues [18] revised the original items to apply to apology proclivity in the context of the parent–child relationship. The PAM-P consists of eight negatively valanced statements (e.g., “I don’t like to admit to my child that I am wrong”). One additional question with a positive valance (“I have a tendency to apologize to my child”) was added to the measure, consistent with prior research on the PAM [24], for a total of nine items. Participants were asked to indicate their agreement with each item using a 7-point Likert scale (1 = “strong disagreement,” 7 = “strong agreement”). The original eight negatively valanced PAM-P items were reverse coded such that higher values indicated a greater proclivity to apologize to their child. The nine PAM-P items were averaged to create an average PAM-P score for each participant. The PAM-P (α = 0.82) demonstrated excellent internal consistency in this study.
Maternal apology effectiveness
Maternal apology effectiveness was measured by coding mothers’ written responses to a fictional vignette, in which a mother becomes angry and yells at her child after falsely accusing them of failing to put their bicycle away. Participants were asked to read the vignette and script a verbatim apology for the scenario. This vignette has been previously described in [17] and is included in the supplementary materials.
Coding scheme
Mothers’ scripted apologies were coded using a directed content analysis approach [25]. Vignette responses were coded for components identified as integral in the construction of an effective apology by prior theoretical and empirical apology research [14, 15, 26, 27]. Specifically, the presence or absence of the following six apology components were coded: 1) remorse/regret (e.g., “I’m sorry for yelling at you”), 2) acknowledgement of unjust or wrong events (e.g., “Yelling at you was wrong”), 3) acknowledgement of victim’s emotions and/or suffering (e.g., “I can see that you’re crying and that I hurt your feelings”), 4) commitment of forbearance (e.g., “I promise that I will work on this in the future”), 5) offer of repair (“What can I do to make you feel better right now?”), and 6) explanation/rationalization of what went wrong (e.g., “I yelled because I had a rough day and jumped to conclusions”). Based on prior literature suggesting that apologies with a greater number of apology elements are more effective [14], components were summed to create a total score reflecting apology effectiveness that was used in data analysis. The coding manual with examples can be found at: https://mfr.osf.io/render?url=https://osf.io/c36dy/?pid=w74cy%26direct%26mode=render%26action=download%26mode=render.
Vignettes were coded by two members of the study team. Interrater reliability was estimated based on a random sample of 33% (n = 63) of the files, using percent agreement and Cohen’s kappa statistic. The average Cohen’s kappa was 0.77, and the average percent agreement was 93.4% across the six components, indicating substantial agreement. However, examining interrater reliability for individual components revealed that the percent agreement and Cohen’s kappa were lower (76.2% and 0.37, respectively) for the “explanation/rationalization of what went wrong” category compared to the other dimensions, and the decision was made to exclude this component for further analyses. After removing this category, the average percent agreement for the five apology components used in subsequent analyses was 96.8% and Cohen’s kappa was 0.87, indicating near perfect agreement. Thus, the final categories included – remorse/regret, acknowledgement of unjust/wrong events, acknowledgement of harm/emotions, commitment of forbearance, and offer of repair. Coding discrepancies were resolved through discussion with the study team.
Difficulties with emotion regulation
Mothers’ difficulties with emotion regulation were measured using the Difficulties in Emotion Regulation Scale (DERS; [28]). The DERS is a 36-item self-report questionnaire in which mothers were asked to rate the frequency that each item relates to them on a 5-point Likert scale (1 = “almost never”, 5 = “almost always”). All items are summed to yield a total score, such that higher scores indicate greater difficulty with emotion regulation. The DERS also yields subscales related to emotional awareness, emotional clarity, regulation strategies, impulse control, goal-directed behavior, and nonacceptance of emotions. For this study, the DERS total score was used in primary analyses to gauge participants’ global emotion regulation difficulties. The DERS yielded good internal consistency (α = 0.96) in this study.
Childhood experiences of parental invalidation
The Invalidating Childhood Environment Scale (ICES; [29]) was used to retrospectively assess mothers’ childhood experiences of parental invalidation. The ICES is a retrospective self-report measure in which participants are asked to rate 14 parental behaviors towards them during childhood by both their mothers and fathers (e.g., “When I was anxious, my parents ignored this”) on a 5-point Likert scale (1 = “never”, 5 = “all the time”). Items are summed to create a total score, such that higher scores indicate higher levels of perceived parental invalidation. The ICES has been shown to demonstrate excellent internal consistency and construct validity samples [29]. In the current sample, the ICES yielded good internal consistency for participants’ mothers (α = 0.92) and fathers (α = 0.93). In this study, both mother and father scores were summed to create a total score for childhood experiences of parental invalidation, as both scores were highly correlated, r = 0.60, p < 0.001.
Data analysis
Pre-Registered Analysis Plan
The hypotheses and analysis plan were pre-registered on the Open Science Framework (OSF) prior to data analysis (available at https://osf.io/8agx5/?view_only=62f618f673e84b8f88d9cf92e12ca4a9). To address our first and second study aim, we calculated descriptive statistics and Pearson r correlation coefficients. To test whether childhood experiences of invalidation were indirectly related to experiences of parent apology characteristics through difficulties with emotion regulation, we examined a path analysis model to estimate both the direct association of childhood experiences of invalidation on the two facets of apology, as well as indirect associations through emotion regulation difficulties. The model was tested with 10,000 bias-corrected bootstrapped samples.
Covariate inclusion
Covariate inclusion decisions were made based on the results of a previous study using the same sample [17]. This preliminary study was distinct in that it examined the measurement of apology effectiveness in detail, the relationship between apology effectiveness components, and their relationships with apology proclivity. Results from this study did not indicate any demographic differences in apology variables, with the exception of apology effectiveness by education F(3, 173) = 2.68, p = 0.05, such that higher education level was related to higher apology effectiveness scores. Differences were examined using age, education status, household income, race, ethnicity, relationship status, and sexual orientation. Based on these prior results, we ran the path model in this study with and without covarying for education level. This did not change statistical conclusions, so we opted to present the model statistics without the education variable for parsimony and consistency with pre-registered analysis plan.
Missing data
Rates of missing data were low (0% for apology variables; 3.8% for DERS; 13.4% for ICES). The higher rate of missing data for the ICES was likely due to its positioning at the end of the study, with participant fatigue contributing to dropout. Full Information Maximum Likelihood was used to handle missing data within path analyses. Data were examined for univariate outliers (defined as exceeding 1.5 multiplied by the interquartile range). Univariate outliers were capped at values corresponding to the lower or upper 5% of the respective distributions. However, these outlier procedures did not affect any statistical conclusions, so we opted to present the results from the raw data.
Statistical software
Data were analyzed using the R package lavaan (Version 0.6.7; [30]). Two-tailed statistical tests with a significance threshold of p < 0.05 were used.