Discrepancies in parents’ perceptions of adult children’s well-being: evidence from mother–father–child triads

ABSTRACT

Most studies of discrepancies in parents’ reports about children’s psychological problems address younger children and psychological problems. The current contribution shifts the focus to adult children and to well-being. In adult intergenerational relationships, knowledge of children’s well-being is more uncertain and there is more room for disagreements to arise, especially in the context of divorce. We analyzed Dutch multi-actor survey data, using a sample of triads of adult children, fathers, and mothers (N = 1,440). Two hypotheses were tested about the origins of discrepancies using structural equation models in which child well-being reports were included of parents and self-reports of children. The analyses supported the notion of relational specificity: when parents have a closer and more harmonious relationship with the child, they evaluate the child more positively than the other parent, after controlling for adult children’s self-reports of well-being. Qualified support was obtained for the depression-distortion hypothesis, with mothers who have higher well-being themselves being more positive about the child. Discrepancies were larger among separated parents than among married parents and parent-stepparent combinations. The conclusion is that parents do not always have similar views of adult children’s well-being and that disagreements are systematic, with bias stemming from the informant and the relationship.

KEYWORDS:

• Informant discrepancies
• intergenerational relations
• divorce
• child well-being

Introduction

Many studies have examined to what extent parents agree on their children’s well-being and psychological problems (Duhig et al., 2000). Most of this research applies to young children and examines psychological problems as measured by scales for depression, anxiety, and internalizing and externalizing problems. The research is mainly clinical and motivated by the goal to improve the quality of the psychological assessment of children for researchers and clinicians. The current contribution shifts the focus from young children to adult children, a case for which parental discrepancies have been studied much less often (Achenbach et al., 2005). Comparing reports of mothers and fathers of the same adult child does not have immediate clinical relevance but is important for several substantive reasons.

First, when children are adults and live independently, they are free to decide how much information they share with their parents and they do not always have incentives to disclose their feelings. Moreover, due to the absence of day-to-day contact, it is more difficult for parents to obtain accurate knowledge of their children than when children live at home. Knowledge of one’s adult child depends on communication with the child and this varies with the amount of contact and the strength of parents’ relationships with adult children (Dykstra et al., 2006). Moreover, uncertainty on the part of parents has increased as a result of the rise in separation and repartnering. These trends imply that parents’ experiences are often heterogeneous within families, with sharp differences in contact, closeness, and support exchange between mothers and fathers (Seltzer & Bianchi, 2013). As a result, there is more room for discrepancies in separated families (Brocker et al., 2023; Pelton & Forehand, 2001) and there is a wider variety of parent types per adult child, with stepparents potentially knowing less about their children than biological parents (Ganong & Coleman, 2017; Thomson, 2014).

Second, parents’ perceptions of adult children’s well-being can have consequences. Accurate perceptions are important in that they are conditions for the degree to which parents provide practical, emotional, and financial support to adult children (Fingerman et al., 2015; Henretta et al., 2018). When parents are aware of the problems of their adult children and in agreement with each other about their evaluations, it is more likely that they will provide the necessary support to them. Both too much and too little support to a child may be perceived as problematic in the relationship (Silverstein et al., 1996). Discrepancies in parents’ evaluations may also have implications for the parents’ own relationship. Many studies have focused on differences between partners in their evaluations of their relationship and have pointed to the negative effects of such differences on communication and marital satisfaction (Kenny & Acitelli, 2001). Similar effects may occur when parents disagree about their children. Different perceptions of parents regarding their children’s well-being may lead to confusion and miscommunication in the relationship. Concerns about children’s emotional problems are a source of stress in the parents' relationship and under such conditions, discrepancies between parents’ perceptions can be an additional strain (Zemp et al., 2017).

Disagreements between informants have been linked to various processes and characteristics, including informants’ own well-being, their social and psychological characteristics, and the context in which they observe others (Berg-Nielsen et al., 2012; Brocker et al., 2023; De Los Reyes et al., 2008; De Los Reyes & Kazdin, 2005; Lohaus et al., 2020). The current study applies existing hypotheses to the case of parents and adult children and tests hypotheses by analyzing unique register-based survey data in which multiple parents and their adult children aged 25–45 participated. In studying the sources of parental discrepancies, structural equation models are developed in which characteristics of mothers, fathers, and adult children are combined to assess differences in parents’ perceptions of the adult child. The data contain an oversample of separated parents and stepparents, thereby also allowing us to explore how family complexity and parental discrepancies are related, a topic about which little is known yet.

Background and hypotheses

Research on reporting discrepancies has been concerned with two different types of problems. One problem lies in discrepancies between self-reports and proxy reports, i.e. what children report about themselves and how other people – fathers, mothers, teachers – see this (Achenbach, 2006; Briggs-Gowan et al., 1996; van der Toorn et al., 2010). Another problem studied in the literature – and the topic of this paper – lies in discrepancies among people rating the same child (Dave et al., 2008; Duhig et al., 2000). In this latter set of studies, several types of raters have been compared, including fathers versus mothers (Duhig et al., 2000), parents versus teachers (Rescorla et al., 2014), and parents versus therapists (Muller et al., 2011).

In many studies, a distinction is made between mean differences on the one hand (directional differences) and correspondence on the other (non-directional differences) (Duhig et al., 2000). Correspondence refers to the degree of correlation between parents’ assessments of the child. Directional differences refer to whether one parent has a more positive or negative assessment of the child than the other parent and to what extent this depends on the characteristics of that parent. The two phenomena do not always coincide. For example, if mothers rate children in a more positive light than fathers (a directional difference), the ratings of mothers and fathers can still be highly correlated in a relative sense.

Differences in parents’ assessments may not only stem from bias but also from incomplete knowledge. Bias is best assessed via directional differences by analyzing which parent is ‘too’ positive or negative about the child. Incomplete knowledge is best assessed via correlations. Incomplete knowledge is a source of measurement error that will reduce the correlation between parents’ assessments. The current study is concerned with bias and studies directional differences between two (or more) (step)parents of the same adult child. For this reason, the hypotheses focus on the possible effects of parents’ individual and relational characteristics on differences in their reports about children’s well-being. Well-being is conceptualized broadly, focusing on adult children’s satisfaction in key life domains.

Theoretically, an important contrast in understanding informant discrepancies lies in whether bias stems from the informant or the relationship. The first and most commonly examined hypothesis argues that parents’ perceptions of their children depend on their psychological well-being. More specifically, it is believed that parents who are more depressed report more negative psychological well-being of children than parents who are less depressed. This hypothesis is directional in that it implies that the more depressed parent of the two is also the most negative about the child. Framed initially as the ‘depression distortion hypothesis’ (Richters, 1992; Richters & Pellegrini, 1989), the hypothesis was later generalized to other aspects of well-being. Several mechanisms have been suggested to explain the hypothesis. The prime reason is believed to be that a negative mood works as a filter in observing the world, leading people to focus more on negative characteristics, not only of other people but also of events and conditions (De Los Reyes et al., 2008).

The hypothesis has been tested often for parent–child discrepancies (for a review, see De Los Reyes et al., 2008), also in the context of separated families (Brocker et al., 2023). For interparental discrepancies, evidence is less widespread and applies only to young children. In a sample of first graders, Mulvaney et al. (2007) showed that a parent’s level of anxiety was negatively associated with that parent’s rating of the child after taking into account the ratings of the other parent and a teacher. In a sample of parents with alcohol problems, Kelley et al. (2017) showed that parent ratings of children’s internalizing problems, after controlling for the other parent’s ratings, were associated with that parent’s depressive symptoms. Similar findings for parental differences were obtained when analyzing teenage children that allowed for including ratings by the child to assess discrepancies (Gartstein et al., 2009; Treutler & Epkins, 2003). Effects were more evident for mothers’ than fathers’ psychological symptoms (Treutler & Epkins, 2003). Most tests of the hypothesis have focused on children living at home, not on adult children.

A competing hypothesis is that bias stems not so much from the informant but from the relationship the informant has with the child. This notion is derived from more general ideas about situational specificity in informant discrepancies (Achenbach et al., 1987). For example, studies of teachers and parents have argued that children behave differently in school than at home and that this is partly responsible for the fact that teachers are more critical than parents about children’s internalizing and externalizing problems (Berg-Nielsen et al., 2012). Applying this idea to two parents of the same child, one can argue that parents differ in the relationship they have with their adult children. One parent may feel closer to a child than the other or have more frequent and pleasant contact. Especially in separated families, studies have shown that parents differ, with the father often having a more negative relationship with the child than the mother (Albertini & Garriga, 2011; Becker & Hank, 2022; King et al., 2004). Moreover, there is evidence that a separation increases interparental differences in contact and closeness compared to married parents (Kalmijn, 2016).

A closer and more harmonious relationship with a child may lead to a more positive view of the child’s well-being for two reasons (De Los Reyes & Kazdin, 2006; Lohaus et al., 2020). First, a good relationship may cause the parent to be more satisfied with the child, which may be projected into a more positive feeling about the child’s well-being. Conversely, a poor relationship with the child may be erroneously attributed to the child’s individual problems. Second, in a good relationship, the more positive aspects of the child’s life may become more salient, leading to a more positive view of the child’s well-being.

Evidence for the hypothesis of relational specificity has come mainly from how parents evaluate younger children and is limited to parent–child differences in reports about well-being (De Los Reyes & Kazdin, 2006; Lohaus et al., 2020). In this design, only one relationship can be used to test the hypothesis (i.e. the parent–child relationship). Comparing informants of the same child yields a stronger design since two relationships are involved (i.e. the mother–child relationship and the father–child relationship). It can then be assessed if there is an association between differences between informants in their relationship with the same child and differences in their perceptions of the child’s well-being. A study of four-year-olds showed that teachers reported more problem behaviours of the child than the parent when the teacher experienced conflict with the child (Berg-Nielsen et al., 2012). In an analysis of children 10–12 years of age, Treutler and Epkins (2003) found that mothers reported fewer internalizing problems than fathers when mothers spent more time with the child.

For the two hypotheses, the causal direction is to some extent ambiguous. For example, there is evidence that parents’ well-being is affected by the problems children experience in their lives and the corresponding degree of well-being of the child (Greenfield & Marks, 2006; Kalmijn & De Graaf, 2012; Offer, 2020). These ‘linked lives’ effects on well-being have been found for both younger and adult children. A similar problem emerges for the effects of the parent–child relationship. Studies have found that emotional problems in children negatively affect the parent–child relationship, especially in separated families (Hawkins et al., 2007). These reverse causal effects may lead to spurious effects of parent well-being and relationship quality on their perceptions of children’s well-being. After all, in this scenario, child well-being is a common cause: it affects parents’ well-being/relationships as well as parents’ perceptions of child well-being.

To minimize the risk such confounding effects, it is important to measure bias as directly as possible. Direct measures of bias are often not available but there are possibilities to use reference points external to the informant. By focusing on adult children, as done in this contribution, children’s own assessments of their well-being can serve as reliable reference point to measure bias. When there is an association between parents’ well-being and their perceptions of children’s well-being, holding constant children’s self-reported well-being, this can be seen as stronger, albeit not conclusive, evidence for a distortion effect. A similar reasoning applies to the hypothesis of relational specificity.

Data and method

Data were used from the survey Parents and Children in the Netherlands (Ouders en Kinderen in Nederland; OKiN). The OKiN was based on a probability sample from the national register of people aged 25–45 who lived in the Netherlands in 2017 (Kalmijn et al., 2018). The sample included a systematic oversample of children with separated parents (Kalmijn et al., 2017). Using the registers, the biological parents of the respondents and the new partners of these parents (if present) were approached with a questionnaire. The children were called ‘anchors’ in the study; the parents were called ‘alters’. Alters were approached directly using the register address, not via the anchor. The total number of participating anchors was 6,485 (response rate of 62%), and the total number of alters was 9,325 (response rate of 38%).

The main reason for the lower response rate of alters was that alters were only approached using CAWI (web interviews). In contrast, anchors were approached using CAWI first and CAPI (personal interviews) in the case of an initial nonresponse. For anchors, we checked mode effects with respect to well-being. CAWI and CAPI respondents did not differ in terms of life satisfaction (t = 0.93), depression (t = 1.56), and self-rated health (t = 0.84). CAWI respondents were significantly more lonely (t = 2.15) but the effect size was small (Cohen’s d = 0.13).

Design

Anchors and alters were approached independently so that there are anchors without alters in the data and alters without anchors. Moreover, the anchor was not always identified in the alter questionnaire since the data collections were not linked; parents reported about a (randomly chosen) subset of their (step)children. Finally, not all anchors have more than one parent in the data and the design requires at least one pair of them. To construct the data, we selected anchors for whom two connected parents of the opposite sex participated. Subsequently, we constructed, for each anchor, pairs of alters who reported about the anchor: (a) the biological mother and father (married or separated), (b) the biological mother and the stepfather, and (c) the biological father and the stepmother. After these selections, we have data on 1,440 alter (parent) pairs and these reported about 1,239 unique (adult child) anchors. Note that an anchor can appear in two or even three parent pairs in the case of separation and stepfamily formation. This is corrected by using robust standard errors.

Using the 1,440 mother-father–child triads, we estimated structural equation models to compare parents while adding self-reported information from children. The self-reported measures of anchor children’s well-being were more elaborate than the proxy reports of parents. Self-reports of adult children are assumed to be more accurate than parent reports and therefore conceptualized as causally prior to parents’ reports of children’s well-being. Moreover, we assume that parents influence each other’s perceptions of the child, for example, via communication and discussions about their child. These considerations lead to a non-recursive causal model, as presented in Figure 1.

Figure 1. Structural equation model for parents’ perceptions of adult children’s well-being and adult children’s self-reported well-being.

We modelled maternal reports of adult child well-being as a function of child well-being and child traits (parameters B_m), the paternal reports of child well-being (parameters C_m), the mother’s relationship with the child (parameters D_m), and the mother’s own well-being (E_m). A similar model was estimated for paternal reports of child well-being (parameters B_f, C_f, D_f, and E_f). The model was able to identify the reciprocal effects of parents’ reports since each parent’s report depended exclusively on the parent’s own characteristics (i.e. the relationship with the child and a set of individual parent control variables). We also estimated a ‘crossing effects’ model where effects of mother’s well-being and mother’s relationship with the child on father’s perceptions were added, as well as the opposite effects (i.e. of father’s well-being and father’s relationship with the child on mother’s perceptions).

The paths from maternal to paternal perceptions and vice versa were constrained to be equal and reflected the influence of parents on each other’s perceptions (C_f = C_m). Including the bidirectional effect in the model is important since it prevents the effect of one parent’s perception on the other’s from being overestimated, as it would have been in separate regression models. In other words, the estimated model corrects for simultaneity bias (Wooldridge, 2020).

Substantively, the model assessed to what extent a parent’s perception was more positive or negative than expected based on how the other parent saw the child and how the children reported about themselves. This setup provides a robust test of the hypotheses since it captures parental differences while holding constant how adult children actually felt. In the crossing effects model, the interpretation is that differences in parents’ relationships with the child are associated with differences in parents’ perceptions of the child, again controlling for the child’s own well-being reports.

Measures - parent data

Given the age range of the children, it was decided not to use conventional mental health measures used in previous studies of reporting differences. Instead, we approached well-being by first defining life domains, following studies on life satisfaction (Weber & Huebner, 2015). Four domains were defined: work/school, partner/family/children, health/well-being, and social ties. For each domain, we asked parents to rate how well the child was doing on a five-point scale: (1) poor, (2) mediocre, (3) not good/not bad, (4) well, (5) very well. If a domain was not applicable, parents could indicate so. The four items were averaged into a scale (α = .83). We also explored domain-specific results but these were broadly similar to the results for the overall scale. Partial missings were addressed using a sequence of regression imputations where other items were used to predict values on items with missing values. Methodological studies have demonstrated that regression imputation without random components performs just as well for item-level missing data compared to more complex multiple imputation methods (Shrive et al., 2006).

The means and standard deviations of the variables are presented in Table 1. Correlations for a subset of parent and child variables are presented in Appendix 1.

Table 1. Descriptive information about the variables in the analyses.

Variable	N	Mean	SD	Min	Max
Anchor data
Child age	1440	33.2	5.4	25	46
Daughter versus son	1440	.527	.499	0	1
Child life satisfaction	1440	4.016	.688	1	5
Child depressive symptoms	1440	1.539	.496	1	4
Child loneliness	1440	1.131	1.556	0	6
Child self-rated health	1440	4.224	.683	1	5
Child income	1434	3.513	1.284	1	5
Child employed	1440	.906	.293	0	1
Child number of sports	1440	1.423	1.15	0	7
Child cultural activities	1440	2.657	1.314	0	5
Child has partner	1440	.787	.409	0	1
Child ever separated	1440	.682	.466	0	1
Child has children	1440	.501	.5	0	1
Child home owner	1440	.653	.476	0	1
Alter data
Mother perception child wellbeing	1439	−.005	1.026	−4.473	1.18
Mother-child closeness	1440	.522	.274	.004	.827
Mother-child conflict	1440	.505	.246	.307	.998
Mother-child symmetry	1440	.489	.245	.051	.67
Mother life satisfaction	1435	−.002	.898	−3.593	1.465
Age mother	1440	60.5	6.2	39	78
Education mother	1434	.345	.476	0	1
Migrant background mother	1440	.084	.278	0	1
Separated mother	1440	.432	.496	0	1
Stepmother	1440	.214	.41	0	1
Alter data
Father perception child wellbeing	1437	.005	.974	−4.76	1.18
Father-child closeness	1440	.478	.265	.004	.827
Father-child conflict	1440	.495	.243	.307	.998
Father-child symmetry	1440	.511	.237	.051	.67
Father life satisfaction	1435	−.002	.898	−3.593	1.465
Age father	1440	63.3	6.3	42	86
Education father	1438	.424	.494	0	1
Migrant background father	1440	.072	.258	0	1
Separated father	1440	.378	.485	0	1
Stepfather	1440	.285	.451	0	1

Note: Adult child-father-mother triads from OKiN 2017.

The type of parent couple was distinguished into three groups: (a) biological parents of the child who were still married or cohabiting (called married for short; n = 438 pairs, 876 parents), (b) biological parents of the child who were divorced or separated (called separated for short; n = 284 pairs, 568 parents), and (c) a biological parent and the current new partner of this parent (called stepparent; n = 718 pairs, 1,436 parents). Keep in mind that not all children lived with the stepparent during youth.

Parent well-being was measured using three items from the Satisfaction with Life Scale of Diener and colleagues (Diener et al., 1985). Response options ranged from 1 = completely agree to 5 = completely disagree. The scale was created based on the mean of the three items. The scale’s reliability was α = .86. The parent–child relationship was assessed based on research on intergenerational solidarity and adult parent–child relations (Dykstra et al., 2006; King, 2006; Roberts & Bengtson, 1990; Silverstein et al., 2010). Contact was measured by asking about the frequency of face-to-face and telephone contact with the child, using seven categories. Closeness was measured by asking the parent how close they felt to the child, using five categories ranging from 1 = not close at all, to 5 = very close. Conflict was assessed by parents’ reports about the frequency of conflict and tension in the relationship (using four categories). Contact, closeness, and conflict were coded into percentile scores and included as linear variables. We added a single-item measure of the extent of symmetry in the relationship. Parents were asked to describe the balance in their relationship to the child using five categories: (1) parent gives more, (2) parent gives a little bit more, (3) equal, (4) child gives a little bit more, and (5) child gives more. The variable was coded as 1 = parent gives more, 2 = parent gives a little bit more, 3 = symmetry or child gives more. Theoretically, the situation in which the parent perceives to give more was believed to be the most relevant. Moreover, only a few parents reported that the child gave more in the relationship (1.3%).

Measures - child data

Two types of measures for anchors (children) were used. First, a set of direct measures of well-being were included. We included the Satisfaction with Life Scale, which was also measured for parents (α = .85). Loneliness was measured using the loneliness scale of De Jong Gierveld and van Tilburg (De Jong Gierveld & Van Tilburg, 2006). Respondents were presented with six items and were asked about the extent to which they applied to their situation (1 = yes, 2 = more or less, and 3 = no). In creating the scale, we took the average across standardized items. The scale’s reliability was α = .82. Self-rated health was measured with the standard question on overall health and was left as a linear variable, coded from 1 = poor to 5 = very good (Miilunpalo et al., 1997). The first three measures were identical to the ones used for parents. Depression was an extra variable and was measured using the 8-item version of the Centre of Epidemiological Studies-Depression Scale. The anchors were asked to rate how well the eight statements described how they felt during the past week on a scale from 1 = rarely or never to 4 = most of the time or always. The scale was calculated by taking the mean of the eight items. The scale’s reliability was α = .86.

Second, a set of indirect measures were included, defined as objective characteristics that are known to affect people’s well-being and that could be used by parents as signals of the child’s well-being in the case of uncertainty or limited communication. The following child characteristics were used: (a) partner status (1 = having a partner, 0 = no partner, regardless of the living situation), (b) married (1 = married, 0 = not married), (c) divorce/separation (1 = ever separated, 0 = never separated), (d) children (1 = child has children, 0 = no children), (e) standardized household income (coded in percentiles, obtained from the population register), (f) employment (1 = child employed, 0 = not employed, obtained from the population register), (g) home ownership (1 = yes, 0 = no), (h) sports participation (the number of sports the child participated in during the past 12 months), (i) cultural activities (a count of five cultural activities the child engaged in during the past 12 months, i.e. attending a classical music concert, attending a popular music concert, reading books, visiting a museum, playing a musical instrument).

Findings

Before analyzing directional differences between parents, we document the level of correspondence between them, using all parent pairs in the data. Correlations between parents’ perceptions of the child’s well-being were r = 0.65 for married parents, r = 0.42 for separated parents, and r = 0.64 for parent-stepparent combinations. These correlations are substantial – except for separated parents – but also leave room for analyzing discrepancies. In Figure 2, the absolute differences between parents are presented, showing that despite agreement for a substantial number of parents, there were also clear differences. Again, there were greater differences between separated parents than between married parents. Interestingly, no overall gender difference was found. The difference in the average well-being ratings between (step)fathers and (step)mothers (‘within children’) was not significant (t = −.38, p = .65).

Figure 2. Differences between parents’ perceptions of adult children’s wellbeing.

Tests of the main hypotheses

Before discussing the SEM models, we discuss linear regression models for fathers’ and mothers’ perceptions in which only anchor child variables were included. These models address a simple question: to what extent do the perceptions of parents correspond to what children feel and do? The results are presented in Table 2. Children’s self-reported well-being had the expected effects on parents’ perceptions of children’s well-being. Effects were found for depressive symptoms, self-rated health, and life satisfaction. No effects were found of children’s loneliness. On top of the effects of these direct measures of child well-being, several anchor behaviours were also associated with parents’ perceptions. Children who were employed and had a partner were ‘rated’ as having a higher level of well-being than children without work and without a partner. Minor positive effects were observed of the child’s cultural activities, sports participation, and home ownership on parents’ perceptions of children’s well-being. The explained variance in parents’ perceptions was 32% for mothers and 30% for fathers. This implies a multiple correlation between all direct and indirect child variables combined, and parents’ reports of child well-being of .56 and .54 respectively. Hence, there was a reasonably good match between parents and children and no difference in this respect between fathers and mothers.

Table 2. OLS regression of parents’ perceptions on anchor child variables.

Anchor child variables	Mothers’ perceptions	Father’s perceptions
Child age	−.010^∼	−.010*
	(−1.96)	(−2.05)
Daughter versus son	.024	−.029
	(.51)	(−.63)
Child life satisfaction	.230**	.175**
	(5.50)	(4.34)
Child depressive symptoms	−.361**	−.336**
	(−5.69)	(−5.49)
Child loneliness	−.017	−.017
	(−.92)	(−.93)
Child self-rated health	.261**	.219**
	(7.06)	(6.14)
Child income	.033	.049*
	(1.56)	(2.37)
Child employed	.495**	.328**
	(5.93)	(4.07)
Child number of sports	.043*	.042*
	(2.06)	(2.12)
Child cultural activities	.026	.053**
	(1.40)	(3.00)
Child has partner	.382**	.511**
	(6.26)	(8.66)
Child ever separated	−.039	−.030
	(−.79)	(−.64)
Child has children	−.002	−.015
	(−.04)	(−.27)
Child home owner	.199**	.106^∼
	(3.33)	(1.83)
Constant	−2.245**	−1.861**
	(−7.07)	(−6.07)
Observations	1433	1431
Adjusted R^2	.318	.296

Note: OKiN 2017 merged anchor-alter data.

^∼ p < 0.10, * p < 0.05, ** p < 0.01

We now turn to the SEM models. The models for mothers’ perceptions are presented in the left-hand panel of Table 3; the models for fathers’ perceptions are presented in the right-hand panel. These models were estimated simultaneously to correctly estimate the mutual effects of parents’ perceptions on each other. We first address the depression distortion hypothesis (hypothesis 1). There were positive effects of parents’ life satisfaction on how they rated their adult children after controlling for the other parent’s rating and the child variables. The effect was significant for mothers and smaller and insignificant for fathers. The BIC for a gender-constrained model was 76025 and this was only slightly worse than the original model with a BIC of 76023 (i.e., a higher BIC indicates a poorer fit). Nonetheless, there was a substantial difference in the magnitude of the two effects: the effect of life satisfaction for fathers was less than half of that of mothers. In the bottom panel of the table, we present the results of the crossing effects model. In this model the effect of mother’s life satisfaction was still statistically significant. The effect of father’s life satisfaction was slightly stronger and now statistically significant, although again considerably weaker than it was for mothers. We conclude that there is confirmation for hypothesis 1, but primarily for mothers.

Table 3. Structural equation models for parents’ perceptions of anchor children’s well-being.

Main model
Mothers’ perception child wellbeing		Fathers’ perception child wellbeing
Stepmother^1	−.003	Stepfather^1	.041
	(−.05)		(.81)
Separated mother^1	.089^∼	Separated father^1	.112*
	(1.86)		(2.38)
Father perception child wellbeing	.210**	Mother perception child wellbeing	.210**
	(14.56)		(14.56)
Mother life satisfaction	.088**	Father life satisfaction	.040
	(3.25)		(1.58)
Mother-child closeness	.804**	Father-child closeness	.874**
	(8.18)		(10.67)
Mother-child conflict	−.334**	Father-child conflict	−.146^∼
	(−3.62)		(−1.71)
Mother-child symmetry	.252**	Father-child symmetry	.276**
	(2.62)		(2.96)
Age mother	−.000	Age father	.005
	(−.05)		(1.05)
Education mother	−.096*	Education father	−.057
	(−2.05)		(−1.33)
Migrant background mother	.215**	Migrant background father	−.019
	(3.13)		(−.25)
Anchor variables	included	Anchor variables	included
Observations	1440
BIC	76023
Chi-2 (versus saturated)	26.0	p = .099
Chi-2 (versus baseline)	1803.2	p < .001
Crossing model
Mothers’ perception child wellbeing		Fathers’ perception child wellbeing
Stepmother^1	−.001	Stepfather^1	.046
	(−.02)		(.90)
Separated mother^1	.084^∼	Separated father^1	.095*
	(1.73)		(1.98)
Father perception child wellbeing	.210**	Mother perception child wellbeing	.210**
	(14.29)		(14.29)
Mother life satisfaction	.090**	Father life satisfaction	.054*
	(3.20)		(2.00)
Mother-child closeness	.815**	Father-child closeness	.895**
	(7.95)		(10.50)
Mother-child conflict	−.325**	Father-child conflict	−.071
	(−3.21)		(−.80)
Mother-child symmetry	.222*	Father-child symmetry	.252**
	(2.23)		(2.62)
Age mother	−.000	Age father	.004
	(−.05)		(.90)
Education mother	−.096*	Education father	−.051
	(−2.04)		(−1.18)
Migrant background mother	.216**	Migrant background father	−.019
	(3.17)		(−.25)
Anchor variables	included	Anchor variables	included
Observations	1440
BIC	76074
Chi-2 (versus saturated)	12.2	p = .275
Chi-2 (versus baseline)	1803.2	p < .001

Note: OKiN 2017 merged anchor-alter data. Anchor child characteristics from Table 2 included. FIML used for missing values. Coefficients and t-values. See text for explanation of crossing model.

¹ Reference is married parent.

^∼ p < 0.10, * p < 0.05, ** p < 0.01

To test the relational specificity hypothesis (hypothesis 2), we looked at the relationship traits of the parents. In line with the hypothesis, we found significant effects of closeness, conflict, and symmetry. Parents were more positive about the child’s well-being compared to the other parent when they felt closer to the child, had less conflict, and regarded the relationship as more symmetrical. These effects were controlled for the effects of the other parent and the child. Although the effects were somewhat different for mothers and fathers, overall, there was little evidence for gender differences. Constraining the effects of the relationship variables to be equal across the models for mothers and fathers improved the fit (the BIC declined from 76023 to 76011), showing that the evidence applied to both genders. The crossing effects model, presented in the bottom of Table 3, yielded roughly the same confirmation of hypothesis 2.

We also observed a strong and significant effect of the parents’ perceptions on each other (b = .210). Important to note is that these effects are net of the effects of the child variables. Without the child variables, the mutual effect of parents’ perceptions was b = .315. Hence, about a third of the association between parents was due to common and underlying characteristics of the child. That there was a remaining effect suggests that there was a tendency of parents to align their evaluations of the child. We explored whether these mutual effects differed depending on the direction of the effect. When allowing the effects to differ between mothers and fathers, we found an effect of .206 from the father’s on the mother’s perception and .213 from the mother’s on the father’s perceptions, a trivial difference. The model’s fit was also poorer (BIC = 76037 versus 76023) when allowing the effects to differ, providing support for the equality constraint.

Evidence for married, separated, and stepparents

In our last set of analyses, we estimated the model separately for different types of parent pairs. We begin with individual models, as in Table 2, where the focus is solely on how well anchor child variables predict parents’ perceptions of the child. Considering the explained variance, we see broadly similar levels of accuracy across types of parents (Table 4). Separated fathers and stepfathers were slightly worse in predicting. For example, the explained variance in perceptions was 27% for separated fathers compared to 33% for separated mothers. For stepfathers the explained variance was 27% compared to 30% for stepmothers. There were some differences in how specific anchor traits predicted perceptions but the differences were small and not systematic.

Table 4. OLS regression for anchor child effects on parents’ perceptions by type of parent couple.

	Married mothers	Married fathers	Separated mothers	Separated fathers	Stepmothers	Stepfathers
Child age	−.011	−.010	−.009	−.013	.004	−.009
	(−1.52)	(−1.27)	(−1.13)	(−1.62)	(.29)	(−.90)
Daughter versus son	−.025	−.046	.113	−.111	−.060	.089
	(−.34)	(−.60)	(1.58)	(−1.54)	(−.55)	(.97)
Child life satisfaction	.350**	.279**	.229**	.040	.071	.205*
	(5.45)	(4.16)	(3.50)	(.61)	(.70)	(2.47)
Child depressive symptoms	−.142	−.308**	−.463**	−.396**	−.417**	−.298*
	(−1.32)	(−2.90)	(−4.75)	(−4.23)	(−3.04)	(−2.29)
Child loneliness	−.054^∼	−.033	−.004	−.010	.010	−.000
	(−1.84)	(−1.09)	(−.15)	(−.35)	(.23)	(−.01)
Child self-rated health	.291**	.151*	.237**	.208**	.297**	.292**
	(5.05)	(2.54)	(4.23)	(3.59)	(3.22)	(3.99)
Child income	.009	.049	.046	.042	.036	.059
	(.26)	(1.35)	(1.45)	(1.33)	(.73)	(1.41)
Child employed	.448**	.288*	.382**	.387**	.662**	.338*
	(3.07)	(1.97)	(3.02)	(3.18)	(3.72)	(2.04)
Child number of sports	.035	.056	.065*	.030	−.011	.041
	(1.06)	(1.64)	(2.02)	(.94)	(−.24)	(1.03)
Child cultural activities	.016	.054^∼	.046	.069*	.027	.038
	(.57)	(1.82)	(1.60)	(2.49)	(.64)	(1.04)
Child has partner	.235*	.561**	.382**	.499**	.527**	.473**
	(2.35)	(5.34)	(4.10)	(5.82)	(3.86)	(3.65)
Child ever separated	.039	−.070	−.117	−.003	.019	−.005
	(.53)	(−.92)	(−1.49)	(−.04)	(.16)	(−.05)
Child has children	.079	.032	−.066	.010	−.064	−.075
	(.87)	(.33)	(−.74)	(.11)	(−.47)	(−.66)
Child home owner	.177^∼	.135	.214*	.070	.211	.127
	(1.80)	(1.34)	(2.38)	(.79)	(1.52)	(1.04)
Constant	−2.835**	−2.058**	−1.926**	−1.084*	−2.551**	−2.476**
	(−5.65)	(−4.01)	(−3.93)	(−2.14)	(−3.33)	(−3.97)
Observations	507	482	620	541	306	408
Adjusted R^2	.313	.332	.334	.267	.302	.271

Note: OKiN 2017 merged anchor-alter data.

^∼ p < 0.10, * p < 0.05, ** p < 0.01

The SEM models are estimated for four types of couples and the results are presented for the two parents in each type of couple next to each other in Table 5: (a) biological mothers and fathers in married families, (b) biological mothers and fathers in separated families, (c) biological mothers and stepfathers in stepfamilies, and (d) biological fathers and stepmothers in stepfamilies. We test two contrasts, between married and separated parents and between married and stepparents. Tests are conducted separately for mothers and fathers and obtained from subgroup analyses in SEM (bottom of Table 5).

Table 5. Structural equation models for parents’ perceptions of children’s well-being by type of parent couple.

Parents’ perception child wellbeing (Y)	(a) Married families		(b) Separated families		(c) Stepfather families		(d) Stepmother families
	Married mothers	Married fathers	Separated mothers	Separated fathers	Biological mother	Stepfather	Biological father	Stepmother
Other parent perception	.222**	.222**	.114**	.114**	.231**	.231**	.254**	.254**
	(8.10)	(8.10)	(4.26)	(4.26)	(8.63)	(8.63)	(9.53)	(9.53)
Parent life satisfaction	.095*	.063	.070	.020	.102*	−.009	.080	.116^∼
	(2.35)	(1.54)	(1.36)	(.39)	(2.21)	(−.18)	(1.52)	(1.95)
Parent-child closeness	.917**	.933**	.368^∼	1.058**	.751**	.852**	.634**	1.103**
	(6.05)	(6.02)	(1.72)	(6.54)	(4.19)	(5.47)	(4.24)	(6.50)
Parent-child conflict	−.318*	−.302*	−.234	.168	−.377*	−.174	−.091	−.466*
	(−2.28)	(−2.12)	(−1.22)	(.90)	(−2.20)	(−1.01)	(−.57)	(−2.16)
Parent-child symmetry	.281^∼	.163	.319	.588**	.233	.229	.196	.126
	(1.92)	(1.02)	(1.61)	(3.19)	(1.43)	(1.23)	(1.14)	(.57)
Age parent	−.007	.002	.015	.000	−.008	.010	−.010	.005
	(−.75)	(.23)	(1.00)	(.04)	(−.71)	(1.63)	(−1.18)	(.86)
Education parent	−.062	.008	−.198*	−.136	−.160^∼	−.085	−.054	.068
	(−.78)	(.11)	(−2.11)	(−1.63)	(−1.69)	(−1.06)	(−.66)	(.74)
Migrant background parent	.195^∼	−.132	.391*	−.151	.259*	.310*	−.099	.175
	(1.77)	(−.95)	(2.17)	(−1.02)	(2.27)	(2.33)	(−.70)	(1.26)
Observations (pairs)	438		284		410		308
Chi-2 (versus saturated)	24.6	p = .04	14.7	p = .40	16.8	p = .27	13.9	p = .45
Chi-2 (versus baseline)	660.4	p < .01	306.8	p < .01	536.4	p < .01	442.4	p < .01
BIC	21235			15436	21717		16616
Wald Chi-2 tests for differences in effects between types	Married versus separated		Married versus stepparent
	Mothers	Fathers	Mothers	Fathers
Parent life satisfaction	0.15	0.44	0.08	1.29
Parent-child closeness	4.40*	0.32	0.67	0.13
Parent-child conflict	0.13	4.01*	0.34	0.33
Parent-child symmetry	0.02	3.04∼	0.34	0.07

Note: OKiN 2017 merged anchor-alter data. Anchor children’s characteristics from Table 2 included. FIML used for missing values.

^∼ p < 0.10, * p < 0.05, ** p < 0.01

In line with the correlational findings, separated parents had a smaller influence on each other’s perceptions than married parents. The difference between the two coefficients was substantial (b = .114 versus .222). The mutual effect was 49% weaker when parents were separated than when parents were married. No difference was noticeable between married parents and parent-stepparent pairs. Clearly, there was much less alignment of perceptions in separated couples.

The depression distortion hypothesis was confirmed for mothers in all types of couples. Even though the effect of mother’s life satisfaction was not always significant, the differences were small in magnitude. The effects were b = .095 for married mothers, b = .070 for separated mothers, b = .102 for mothers in stepfather families, and b = .116 for stepmothers. The tests reveal no significant differences in the life satisfaction effects between types of mothers. For fathers, the effect of life satisfaction was not significant in any type of couple, confirming that the evidence for the hypothesis was weak for fathers. The tests were insignificant for the differences between types of fathers in the effect of life satisfaction.

The hypothesis of relational specificity was confirmed again in the subgroup analysis. For the perceptions of mothers, there were a few differences, with closeness to the child being less important for separated mothers than for married mothers. This difference was statistically significant. For the effects of the other relationship characteristics, there were small differences in magnitude and tests for differences were insignificant. For fathers, there were more differences. Conflict was important for married fathers and not for separated fathers and this difference was significant. Closeness, on the other hand, was important for fathers in all types of couples. Interestingly, symmetry was only important for separated fathers and this was significant when compared to married fathers. Since separated fathers often have less contact with their children than they prefer (Kalmijn, 2022), the issue of (a)symmetry is likely to be more salient in this group. Despite these differences, in all groups of parents there were significant effects of relationship characteristics, in line with the hypothesis.

Discussion and conclusion

Parents agree to a large extent about the well-being of their adult children. Still, there are discrepancies between them, and these are systematic in nature. Using a large national sample of mothers, fathers, and adult children, we showed that mothers – but not fathers – with lower levels of well-being had a more negative perception of their children’s well-being. Poor well-being distorts the views mothers have of their children. Theoretically, this can be explained by various cognitive processes, such as more focus on negative signals when well-being is low and more sensitivity to the negative experiences and expressions of others. This finding is in line with psychological studies of young children. Our evidence generalizes the evidence to adult children, in which case there can be more uncertainty about, and heterogeneity in how parents see their children. Moreover, the evidence applies to general well-being and shows that depression distortion does not apply exclusively to more clinical problems of children.

We found considerable evidence that the evaluation of a child depends on the relationship in which the parent observes the child. These effects are in line with the hypothesis of situational specificity, in this case, applied to the relationship instead of the broader context of observation, such as the parental home or the classroom. When the parent has a closer, more harmonious, and symmetric relationship with the child, there will be more focus on the positive sides of the child’s life, and the child may reveal more of their positive side. This change in focus can lead to a more positive perception of the child’s well-being. Moreover, when a parent has a negative relationship with the child, this may be attributed to the child, leading to a more critical view of how the child is doing.

An important caveat is that parent–child relationship quality and parent well-being may also be the consequence of child well-being. Emotional problems on the part of children may put a strain on the parent and the relationship (Greenfield & Marks, 2006; Kalmijn & De Graaf, 2012; Offer, 2020). The observed effects may therefore in part be spurious. Child well-being may affect both parent well-being and parents’ perceptions, creating a spurious correlation between parents’ well-being and their perceptions of the child. Similarly, child well-being may simultaneously affect the relationship and parents’ perceptions of children’s well-being, again creating a spurious relationship. The claim in this paper is not about effects at the individual level, however, but about effects at the family level: differences in parent–child relationship quality are associated with differences in perceptions of child well-being, even after controlling for adult children’s own report about their well-being. Assuming that children’s reports about their well-being are the most accurate, our design provides a more direct measure of parental bias. In a sense, the outcome in our model is the degree to which the perception of the parent deviates from the children’s actual well-being. We find that this deviation is affected by parents’ individual and relational characteristics. Although a triadic approach is a step forward to eliminating spurious effects, longitudinal data are still needed to estimate true causal effects.

A shift was made in this paper from childhood, which most research is concerned about, to adulthood, which has been studied more rarely. The case of adult children is important for several reasons. A substantive reason is that there is more uncertainty about children’s well-being when they are adults and live independently. Parents remain concerned about their children’s well-being long after they leave home (Milkie et al., 2008), so this uncertainty is an important issue in families. Moreover, making good inferences about well-being in adulthood is important for providing children with the right amount of emotional, social, and practical support. Especially in the case of separation and repartnering, there is considerable variation in what parents know about their children. A methodological advantage is that in adulthood, reports of children about their own well-being may be a more mature and more valid reference point for comparisons than reports of younger children (Richters, 1992; van der Toorn et al., 2010). This allowed us to estimate models in which elaborate (direct and indirect) measures of children’s well-being predicted parents’ perceptions.

The present study made a number of additional contributions to past research. Many previous studies on discrepancies were based on young children and on small, often select samples. The current paper analyzed a large probability sample with data from three independent sources, thereby strengthening the evidence, both in terms of the amount of agreement (descriptive evidence) and hypotheses about the origins of disagreement (explanatory evidence). A second contribution is that only a few studies have so far tested the role of the relationship with the child for bias in parents’ perceptions (Lohaus et al., 2020). Moreover, these earlier studies primarily measured aspects of communication, acceptance, and time spent with the child and did not compare relationships across parents (e.g. Treutler & Epkins, 2003). Our evidence strengthens the conclusions from these earlier studies using a broader set of relationship traits in a triadic design where parents’ relationships with their children are compared.

The study used a large oversample of separated parents, thereby enabling an analysis of how parents differ in their perceptions of the child when they are no longer together. The degree of correspondence of parents was shown to be substantially smaller in separated couples than in married couples, and the mutual influence parents have on each other’s perceptions, after controlling for children’s characteristics, was smaller. These findings point to another yet undocumented effect of separation on families. Not only does parental separation lead to changes in well-being and heterogeneity in parent–child relationships, but parental separation also changes the way family members perceive the same ‘others’. This finding echoes psychological research on divorce which has suggested that there are substantial discrepancies in separated parents’ perceptions of the causes and consequences of their divorce (Gray & Cohen Siver, 1990; Mikelson, 2008). Or to put it differently, the family provides a context in which perceptions are aligned. Without that context, or with that context broken, people’s perceptions of the same stimuli may begin to diverge.

Gender differences appeared to be small. Mothers and fathers did not differ in their mean ratings of the child and the determinants of their ratings were broadly similar. However, some evidence was present for stronger effects of well-being for mothers, especially in separated families. One previous study also found a gender difference among married parents, with stronger evidence for depression distortion for mothers’ than fathers’ psychological symptoms in the case of younger children (Treutler & Epkins, 2003). Still. the total correlation between all child-reported traits and parents’ perceptions of well-being was the same for mothers and fathers, showing that there was no gender difference in how well parents can ‘predict’ how their children are doing.

We close with a number of limitations of the study and prospects for further study. First, the study was done in one country and one historical context. The Netherlands is a relatively normal example of a European country that has experienced the Second Demographic Transition (Lesthaeghe, 2014). Adult intergenerational relations are relatively strong but in a normative sense, the country can be situated at the individualized pole of the individualism-familialism continuum (Hank, 2007; Reher, 1998). It remains to be seen to what extent the current findings also apply to other, more familialistic contexts or to countries with higher levels of divorce such as the US and Australia. Second, our study was cross-sectional, reducing opportunities to examine changes in well-being, relationships, and perceptions. Obviously, such a design would lead to stronger causal inferences but it is difficult to collect panel data among complete triads, let alone data on triads over longer periods of time where there is sufficient change in well-being. Third, our main outcome variables referred to general levels of well-being and not to psychopathology. Measures such as depression, anxiety, and other more serious mental health issues were not present, at least not from a parents’ perspective. Studies on younger children often include such measures and it would be important to also study these for adult children.

Ethical clearance

Ethical clearance was provided by the Ethical Advisory Board of the University of Amsterdam and the European Research Council. The ERC provided the funding for collecting the data. The data can be used for free and are available in the public domain at the Dutch scientific data archive DANS-EASY (https://easy.dans.knaw.nl/ui/home).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The survey was conducted as part of a larger project on Family Complexity funded by the European Research Council under the Horizon 2020 program [grant no 669334]. The OKiN was developed, designed and executed by a collaborative team of researchers from the University of Amsterdam (UvA) and Statistics Netherlands (CBS). The data are available at DANS-KNAW , the national data archive of the Netherlands.

Appendix 1.

Correlation matrix of all variables except social-demographic traits

Table

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)
1. Life satisfaction (K)	1.000
2. CESD (K)	−0.548	1.000
3. Loneliness (K)	−0.432	0.587	1.000
4. Self-rated health (K)	0.354	−0.355	−0.174	1.000
5. Perceived well-being (M)	0.422	−0.407	−0.309	0.345	1.000
6. Closeness (M)	0.068	−0.051	−0.098	−0.033	0.300	1.000
7. Conflict (M)	−0.166	0.130	0.163	−0.097	−0.259	−0.145	1.000
8. Symmetry (M)	0.150	−0.136	−0.083	0.063	0.217	0.163	−0.297	1.000
9. Life satisfaction (M)	0.105	−0.049	−0.075	0.073	0.145	0.063	−0.122	0.028	1.000
10. Perceived well-being (F)	0.390	−0.383	−0.295	0.320	0.606	0.139	−0.217	0.179	0.056	1.000
11. Closeness (F)	0.110	−0.087	−0.111	−0.020	0.175	0.287	−0.080	0.149	0.087	0.326	1.000
12. Conflict (F)	−0.152	0.104	0.112	−0.057	−0.166	−0.075	0.377	−0.139	−0.062	−0.193	−0.200	1.000
13. Symmetry (F)	0.126	−0.108	−0.066	0.022	0.159	0.103	−0.139	0.306	−0.020	0.183	0.145	−0.207	1.000
14. Life satisfaction (F)	0.105	−0.049	−0.075	0.073	0.145	0.063	−0.122	0.028	1.000	0.056	0.087	−0.062	−0.020	1.000

Note: K is anchor child, M is alter (step)mother, F is alter (step)father. N = 1,440.