Cheng, S-T., Li, K-K., Leung, E.M.F., & Chan, A.C.M. (2011).

Social exchanges and subjective well-being: do sources of positive and negative exchanges matter? The Journals of Gerontology, Series B:
Psychological Sciences and Social Sciences, 66(6), 708718, doi:10.1093/geronb/gbr061. Advance Access published on July 9, 2011.

Social Exchanges and Subjective Well-being: Do Sources

of Positive and Negative Exchanges Matter?
Sheung-Tak Cheng,1 Kin-Kit Li,2 Edward M. F. Leung,3 and Alfred C. M. Chan4

1Department of Psychological Studies, Hong Kong Institute of Education, China.

2Department of Applied Social Studies, City University of Hong Kong, China.
3United Christian Hospital, Hong Kong, China.
4Asia-Pacific Institute of Ageing Studies, Lingnan University, Hong Kong, China.

Objectives. To decompose the effects of positive and negative social exchanges on well-being in terms of relationship
type (vertically extended family, horizontally extended family, or nonfamily) and subjective closeness (close vs. peripheral).

Methods. One thousand and five Chinese older adults rated each network member on positive and negative exchanges,
which were aggregated for each relationship type and closeness category. Regression analyses estimated the influences of
positive and negative exchanges on well-being, controlling for network size, health, and demographic factors.

Results. Social exchanges with close and peripheral vertical family members as well as close horizontal family mem-
bers were associated with well-being, whereas exchanges with nonkin did not contribute independent effects. These
results were similar for both positive and negative exchanges.

Discussion. Well-being is determined not just by social exchanges but also by where they come from. In this regard,
the vertical family, the horizontal family, and the nonfamily represent a hierarchy of preference for Chinese older adults,
which, to some extent, reflects the influence of familism.

Key Words: Chinese older adultsKinshipNegative exchangesPositive exchangesSocial networkSubjective

well-being.

S OCIAL relationships are not always supportive and

pleasant as they can be problematic at times (Rook,
1984). One perspective, the domain-specific effect model
may be more adaptive socially and physiologically (e.g.,
Skowronski & Carlston, 1989; Taylor, 1991). Moreover,
whereas positive and supportive behaviors from others may
(Ingersoll-Dayton, Morgan, & Antonucci, 1997), suggests or may not be viewed favorably, negative interactions are
that positive exchanges predict positive well-being, and always seen as undesirable (Newsom, Rook, Nishishiba,
negative exchanges predict negative well-being. However, Sorkin, & Mahan, 2005), leading to a more consistent rela-
the literature has generally supported a negativity effect tionship between negative exchanges and well-being. Despite
model in which negative exchanges, though relatively infre- this burgeoning literature, an issue that remains unanswered
quent (Rook, 2001), are more predictive of subjective well- is whether the influences of positive and negative exchanges
being when they occur than positive exchanges (Newsom, depend on where they come from. In the present studies,
Nishishiba, Morgan, & Rook, 2003; Rook, 2001), although we consider two dimensions of the source of exchange:
other studies have found opposite results (Okun & Keith, (a) relationship type and (b) degree of closeness.
1998), including one study with a large sample of Chinese
older adults (Li & Liang, 2007). For example, in a sample of
older Americans, although domain-specific effects were found Background
cross-sectionally, negative, but not positive, exchanges pre-
dicted positive (PA) and negative affect (NA) longitudinally, Social Exchanges in Family and Friends
suggesting that negative exchanges may have more enduring Although no one would doubt that both family and
effects than positive exchanges (Newsom et al., 2003). friends play important support functions (Adams &
The disproportionate impact on well-being due to the Blieszner, 1995), some scholars have taken the view that
negative aspects of social interactions may be understood friends may be more contributive to well-being in older
from a few theoretical propositions. Family and friends adults than the family. A major reason has to do with the
are expected to be supportive. Hence, negative exchanges nature of activities spent with family and friends. Using
may exert a greater impact because of the violation of this experiences sampling method, Larson, Mannell, and Zuzanek
expectation (Rook & Pietromonaco, 1987). Others have (1986) found that the times spent with friends, which were
suggested that greater arousal, cognitive processing, and characterized by activities of common interests and sponta-
emotional responses to negative interactions or life events neity, were more conducive to positive affective states in
708 The Author 2011. Published by Oxford University Press on behalf of The Gerontological Society of America.
All rights reserved. For permissions, please e-mail: [email protected].
Received December 11, 2009; Accepted May 7, 2011
Decision Editor: Rosemary Blieszner, PhD
 POSITIVE AND NEGATIVE EXCHANGES 709

Canadian older adults than the times spent, usually in mun- fame; the feelings of familial unity (being one), belong-
dane and routine daily activities, with family members. ingness, concern and love, glory, responsibility, and safety;
Similarly, Merz and Huxhold (2010) found that, among and the tendencies to engage in such behaviors as produc-
older Germans, instrumental support from kin undermined ing offspring, interdependence, forbearance, modesty, con-
well-being, whereas instrumental support from nonkin en- formity, striving for family, respect for seniority, and
hanced well-being, although the hampering effect of instru- in-group favoritism. The widespread endorsement of these
mental kin support disappeared after controlling for health. cognitive, affective, and behavioral attributes suggests that
Nonkin instrumental support might be particularly well- the familynonfamily divide among Chinese people may
being enhancing in this German sample because of its be more prominent than those among Westerners, although
voluntary rather than its obligatory nature. this is an empirical question.
Another potential reason for friends being more well- Furthermore, it is important to note that whereas in the
being enhancing than family members is that they are less West, the notion of the family refers usually to the immedi-
likely to be the main source of negative interactions in daily ate family, in Chinese and other Asian societies, the
life. Compared with family relationships, friendships are extended family is typically included. The demolition of the
not bound by prescribed roles and are grounded on mutual- extended family in Chinese and Asian societies may be
ity and reciprocity. Friendships loaded with negative ex- largely structural but not functional. Recently, Cheng, Lee,
changes would probably be discontinued. Indeed, research Chan, Leung, and Lee (2009) argued that not only are
shows that negative interactions are more frequent among immediate kin important for the Chinese people, distant kin
family members than among friends, whether in Western or also play significant roles, particularly for those without
Asian societies (Akiyama, Antonucci, Takahashi, & Langfahl, immediate kin. They found that the vertically extended family
2003; Krause & Rook, 2003). Moreover, a longitudinal (including parents, spouse, children, children-in-law, grand-
study spanning over 10 years in a large sample of U.S. children, etc.), the horizontally extended family (siblings,
adults found that the degree of negativity decreased in cousins, distant in-laws, etc.), and the nonfamily (friends,
friendship over time, owing to the discontinuation of prob- neighbors, etc.) represented three distinct social circles in
lematic relationships (65% of respondents reported a change the lives of Chinese older adults. Whereas contact and sup-
in best friend from T1 to T2), but negative exchanges with port exchanges were highly correlated within each of the
spouse or partner increased over time (except for the minor- three relationship categories, they were uncorrelated across
ity who changed spouse or partner), and those with children categories. Thus, the study of social exchanges and well-
remained stable (Birditt, Jackey, & Antonucci, 2009). These being needs to be placed within the broader context of the
findings suggest that, in previous studies that examined neg- extended family. To the best of our knowledge, this is the
ative interactions and well-being, the effects might have first study that directly compares distant kin (horizontally
been mainly driven by family relationships than by friend- extended family) to immediate kin (vertically extended
ships. Nevertheless, this possibility was never directly family) in terms of their relationships to subjective well-
tested in the literature. being in older adults. Because Chinese people have greater
or more established expectations for family than for friends
(e.g., filial expectations for children; Cheng & Chan, 2006),
Family and Friends in the Chinese Context we expect that the effects of positive and negative exchanges
Not all Western findings show that friends are more con- will be strongest if they come from immediate kin, less
tributory to subjective well-being than the family. Thomp- strong if coming from distant kin, and weakest when coming
son and Heller (1990) found that perceived and structural from nonkin.
support from the family, but not those from friends, were In this connection, it is important to point out that an ear-
predictive of psychological well-being in a U.S. sample of lier study (Cheng et al., 2009), though examining similar
older women, whereas friendship ties were important only issues, did not provide a direct comparison of the effects
for those with weak family ties, suggesting a possible com- of the three relationship categories. That study found that
pensatory function for friends (Cantor, 1979; Rook & Chinese older adults with family-focused networks (charac-
Schuster, 1996). Research studies that focused on childless terized by high contact and high support exchanges with the
older adults in European countries have also shown similar vertical family as well as low contact but high support
compensatory effects (Albertini & Kohli, 2009; Wenger, exchanges with the horizontal family) or distant family
2009). This compensatory function may be more relevant networks (characterized by high support exchanges with the
for the Chinese person who has a tradition of espousing horizontal family) were comparable to those with friend-
familism. Familism refers to the extent to which people focused networks across a variety of well-being indicators.
identify themselves with their family (Bardis, 1959). Ac- However, network types are constellations of support re-
cording to Yang (2006, p. 300), familism in the Chinese sources (i.e., a distant family network may also include
context is characterized by the emphases on ones own some immediate kin and friends and so on), they could
familys prolongation, harmony, solidarity, wealth, and not isolate the effects coming from different sources. It is
710 CHENG ET AL.

therefore important to find ways by which the effects of decrease from vertical family to horizontal family, and further
different relationship types can be compared within the to nonfamily, across levels of closeness.
same person.
Methods
Social Exchanges by Degree of Closeness
Another related issue, also untapped by the literature, is Participants and Procedure
whether the effects of positive and negative exchanges Participants were 1,005 community-dwelling older per-
depend on whether they occur in interactions with close sons in Hong Kong (Mage = 72.2 years [SD = 7.15, range =
or peripheral network members. A common approach by 5998]; 51% women) who were recruited on a convenience
gerontologists defines close network members as those who basis by placing advertisements in social service agencies
are considered so close that it would be hard to imagine or through referrals by agency staff. Participants were inter-
life without them (Lang & Carstensen, 1994, p. 317)that viewed face to face either at their home or at a social center.
is, the inner circle in the social convoy model (Kahn & All scored 20 on the Mini-Mental State Examination
Antonucci, 1980). In contrast, peripheral network members (Folstein, Folstein, & McHugh, 1975) and provided
are taken as those placed in the middle (very important but informed consent to participate. This study was approved
less close) and the outer (even less close but still important) by the Ethics Subcommittee of the Research Committee of
circles (Lang & Carstensen, 1994). This being the case, it is the City University of Hong Kong.
likely that the close network members, so essential to self, Mirroring the general older population in Hong Kong
would have more impact than peripheral members on ones (Census and Statistics Department, 2007), this sample was
well-being, whether the interactions were positive or nega- a low-income group with low educational level (36% with
tive. On the other hand, it is possible that negative interac- no formal education, 38% primary education, 21% second-
tions with close network members are more easily forgiven ary education, and 5% tertiary). Most of the participants
(Tse & Cheng, 2006) in order to maintain emotionally (84%) reported a monthly income of less than 6,000 Hong
meaningful relationships in old age (Cheng & Yim, 2008). Kong dollars (about US$770), and 42% of the participants
Furthermore, caring behaviors by someone less close may were recipients of the Comprehensive Social Security
be felt with more gratitude than the same behaviors from a Assistance (CSSA) Scheme, a subsistence support provided
very close person because one feels less deserving of the by the government to low-income persons. About half of the
former (see Emmons & McCullough, 2004). Thus, it is not participants lived alone (51%). Those divorced, separated,
clear in what way closeness would influence the relation- and single were oversampled. Participants had an average
ships between social exchanges (especially negative ex- of 2.45 children (SD = 2.16). More details about the sample
changes) and well-being. The issue is further complicated can be found in Cheng and colleagues (2009).
by the fact that universally kin are, in general, regarded as
closer than nonkin (Neyer & Lang, 2003). Hence, the only Measures
way to disentangle the effects of relationship type and
closeness is to study the relative impact of the vertical family, Degree of closeness and relationship types.Participants
the horizontal family, and the nonfamily broken down by responded to the social convoy questionnaire by Kahn and
levels of closeness. Antonucci (1980), which shows three concentric circles
drawn around a center labeled me. They were asked to
place network members into the circles depending on how
The Study important these persons were in their lives. Following Lang
In this study, we tested the simultaneous associations of and Carstensen (1994), those placed in the inner circle (so
positive and negative exchanges with subjective well-being, important that it is difficult to imagine life without them)
broken down by relationship type and closeness in a large were classified as close network members and those in the
sample of Hong Kong Chinese older adults. Consistent with middle (very important but not as close) and outer (close
the literature, we hypothesized that positive exchanges enough and important enough but have not yet been included)
enhance well-being, whereas negative exchanges undermine circles as peripheral network members. Participants also
well-being, across relationship types and levels of close- reported their relationships with each of the network
ness. However, because of previous inconsistent findings, members, which were used to classify network members
we did not hypothesize the relative effects of positive and into vertically extended family (parents, spouse, children,
negative exchanges but would nonetheless test whether children-in-law, grandchildren, and great grandchildren),
their associations with well-being are equivalent. Addition- horizontally extended family (including siblings, cousins,
ally, for reasons mentioned earlier, we did not make a spe- and all other distant kin and in-laws), and nonfamily
cific hypothesis concerning the effect of closensss, but we (friends, neighbors, and other nonfamily network mem-
expected that the effects of positive and negative exchanges bers). Hence, six subgroups including close vertical family,
 POSITIVE AND NEGATIVE EXCHANGES 711

peripheral vertical family, close horizontal family, periph- Covariates.Age, sex (1 = female and 0 = male), mar-
eral horizontal family, close nonfamily, and peripheral non- riage (1 = married and 0 = not married), socioeconomic
family were created by combining these two dimensions. status (SES; a composite score combining the standardized
scores of educational level and whether the participant
Positive and negative exchanges.Participants provided received CSSA, with higher scores indicating higher SES
ratings on a 5-point scale of 1 (never) to 5 (always) the levels), and living arrangement (1 = living alone and 0 = not
degree to which he or she had received emotional support living alone) were included as covariates. In addition, number
(four items), instrumental support (four items), and com- of chronic illnesses (range = 021), instrumental activities
panionship (two items) from each network member. The of daily living (IADL; Lawton & Brody, 1969), and network
participants also rated their provision of emotional (two sizes (broken down by relationship type and closeness)
items) and instrumental support (two items) to each net- were also statistically controlled. IADL was coded as 0 (no
work member. Following Rook (2001), these 14 items impairment) and 1 (some impairment) because only 7% of
were summed to produce a measure of positive exchanges the participants reported some degree of IADL impairment.
(a = .88). Negative exchanges were measured by four items
(conflicts, excessive demands, annoying attitudes and be-
haviors, and unfulfilled promises) on the same 5-point scale Data Analysis
(a = .71). Both measures were aggregated and averaged In order to examine the relative effects of different types
across all network members by relationship type (horizontal, and sources of social exchanges, positive and negative
vertical, and nonfamily) and level of closeness (close and exchanges were each broken down by two dimensions,
peripheral). namely relationship type (3) and closeness (2), which
resulted in six measures each for positive and negative
Subjective well-being.Referring to peoples evalua- exchanges. These 12 measures were used to predict simul-
tions of their lives and encompasses both cognitive judg- taneously each of the 4 measures of well-being by multiple
ments of satisfaction and affective appraisals of moods and regression analyses with robust standard errors (StataCorp,
emotions (Kesebir & Diener, 2010, p.18), subjective well- 2009), which computes the HuberWhite estimates of the
being is usually operationalized in terms of life satisfaction, standard errors that are robust to the violation of normality
PA, and lack of NA (Kesebir & Diener, 2010). Because NA and homoscedasticity assumptions (White, 1980). This
is characteristically low in older adults (Cheng, 2004), estimator was deemed the most appropriate because the dis-
Cheng and Chan (2006) introduced depressive mood as a tributions of measures of negative exchanges were skewed,
fourth measure so as to provide an alternative measure of which might lead to heterogeneity of residual variance.
negative well-being as well as a more balanced assessment Age, sex, marriage, SES, living arrangement, number of
of subjective well-being as a whole. Hence, we measured chronic illnesses, IADL, and network sizes were controlled
subjective well-being by (a) the Satisfaction With Life Scale in the analyses.
(Diener, Emmons, Larsen, & Griffin, 1985), rated from 1 For the measures of positive and negative exchanges, the
(strongly disagree) to 7 (strongly agree; a = .78), (b) the PA minimum possible value was replaced (i.e., 1 = never) if
(six items; a = .87) and NA (six items; a = .85) subscales of there was no network member in that category (the numbers
the Chinese Affect Scale (Cheng, 2004), rated on a 5-point of cases with zero network members were 323 for close ver-
scale of 1 (rarely) to 5 (often) against the past week, and tical family, 789 for peripheral vertical family, 662 for close
(c) a 4-item version of the Geriatric Depression Scale (rated horizontal family, 563 for peripheral horizontal family, 869
on a yes/no basis; a = .78) that has been shown to be equally for close nonfamily, and 295 for peripheral nonfamily).
predictive of psychiatrist ratings of depression compared Those who scored 0 on a network size measure also scored
with the 15- and 30-item versions (Cheng & Chan, 2004; 1 in the corresponding positive and negative exchange mea-
(Cheng, Yu, et al., 2010). A maximum likelihood confirmatory sures. Thus, the effects of the missing value replacement on
factor analysis (correcting for nonnormality) was conducted the well-being variables would be eliminated by the pres-
on the covariance matrix of all these items as indicators for ence of the network size variables, whereas the missing
four first-order factors (life satisfaction, PA, NA, and value replacement allowed the use of the whole sample for
depression), which were then specified to be the result of examining the simultaneous effects of positive and negative
a common second-order factor of subjective well-being. exchanges across relationship type and closeness cate-
Results showed that the items from the four measures gories. Predictors and outcomes were standardized before
tapped a common construct of well-being: SatorraBentler entering the regression models.
scaled c2(179) = 594.49, p < .001; comparative fit index = In addition, to provide more empirical evidence for the
.98; nonnormed fit index = .97; standardized root mean negativity effect or the domain-specific effect model, the
square residuals = .05; and root mean square error of overall effects of positive and negative exchanges were
approximation = .05. All factor loadings (first- and second- compared. The difference in the overall effects of positive
order) were .40. and negative exchanges was tested by comparing models
712 CHENG ET AL.

with and without constraints on their regression coefficients tent across the various measures of well-being. Finally,
using an F test (Chatterjee & Hadi, 2006, p. 70; see Appendix). there were no significant effects coming from the nonfamily.
An equality constraint was applied such that the sum of the
six coefficients (bs) regarding positive exchanges was equal Sensitivity checks.Sensitivity analyses were conducted
to the sum of the six coefficients regarding negative ex- to examine whether the results were affected by missing
changes (the signs of which were reversed) for each of the value replacement (i.e., by 1 = never). Because no partici-
measures of well-being. The standardization of the variables pants have reported all six categories of social relationships,
created a common metric for combining the coefficients. a full analysis without value replacement was not possible.
Alternatively, separate analyses were conducted with posi-
tive and negative exchanges aggregated (a) within relation-
Results ship types only (without regard to closeness so as to ensure
a subsample without zero network member in any category;
Descriptive Results n = 351) and (b) within closeness categories only (without
Across the 1,005 participants, 11,200 network members regard to relationship type; n = 829). After controlling for
were reported, including 3,986 vertical family members the covariates, the regression coefficients were slightly
(78% close), 3,234 horizontal family members (38% attenuated as compared with the analysis based on the
close), and 3,980 nonkin (8% close). At the network mem- whole sample, probably due to decreased variability in the
ber level, relationship type and closeness were strongly variables. However, statistical significance and the relative
related, c2(2) = 3,951.52, p < .001; Cramers V = .59, indicating effect size of the predictors remained essentially the same.
that vertical family members were more likely considered Thus, the results were not affected by missing value
as close and nonkin as peripheral. This is consistent with replacement.
cross-cultural findings (e.g., Japan, Germany, and the
Supplementary analyses.To strengthen the interpreta-
United States) on the social network composition of older
tion of the findings, two sets of supplementary analyses
adults (Antonucci & Akiyama, 2004; Neyer & Lang, 2003).
were conducted. First, it was not clear to what extent the
Descriptive statistics and correlations of the major study
findings were influenced by the characteristics of the sam-
variables are presented in Table 1. Correlations among the
ple (e.g., marital status and health status). To examine
social exchange variables, both within and across relation-
whether the effects of social exchange might be moderated
ship types and levels of closeness, were weak, ranging from
by health and sociodemographic factors, we extended the
.28 to .28. However, the correlations between network size
preceding analyses by including interaction terms between
and positive exchanges were moderate (rs = .32.63, ps <
social exchange variables and age, sex, marriage, SES, living
.001). Thus, the variance inflation factors (VIFs) were
arrangement, chronic illnesses, and IADL. Each moderator,
examined in regression analyses for possible multicollinearity
being associated with 12 interaction terms, was tested in a
problems.
separate regression model. Across all the analyses, the only
significant interactions were found in negative exchanges
Well-being as a Function of the Source of Social with peripheral nonfamily members in which the relationship
Exchanges between negative exchanges and well-being was generally
Results of the regression analyses are presented in Table 2. more negative in older participants and women. Both of these
Altogether, the predictors accounted for 12%22% of the effects were small. Thus, the study findings generally held
variances in the measures of well-being. The VIFs of the regardless of health and sociodemographic characteristics.
predictors ranged from 1.03 to 1.89, which were well within Second, the findings might be due, in part, to the fact that
acceptable limits. The standardized regression coefficients most of the close network members were also vertical family
revealed that older and female participants reported better members. Thus, the contributions of horizontal family and
well-being. As expected, fewer chronic illnesses, no diffi- nonfamily members might have been underestimated when
culty in IADL, and more network members (especially of their effects were estimated together with the greater num-
peripheral nonfamily and close vertical family) were associ- ber of vertical family members. Therefore, the regression
ated with better well-being. Regarding positive and negative analyses were repeated in a subsample of those without any
exchanges, those from close vertical family and close hori- vertical family member. Of this subsample (n = 295), 984
zontal family members were associated with well-being in horizontal family members (64% close) and 1,158 nonkin
the predicted directions. Moreover, positive and negative (21% close) were reported. Controlling for the covariates,
exchanges with peripheral vertical family members were positive and negative exchanges with close horizontal family
related to well-being, whereas only positive exchanges (but members were associated with virtually all measures of
not negative exchanges) with peripheral horizontal family well-being, and negative exchanges with peripheral hori-
members were related to well-being. The effects of social zontal family members were also associated with PA (Table 3).
exchanges within the peripheral network were less consis- These results were similar to those reported in Table 2 for
 Table 1. Descriptive Statistics of Well-being Measures and Social Exchanges (broken down by relationship category and closeness; N = 1,005)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
1. Depression
2. Life satisfaction .51
3. Positive affect .62 .59
4. Negative affect .69 .49 .58
5. Net. size (VF-C) .10 .20 .17 .07
6. Net. size (VF-P) .05 .07 .04 .08 .01
7. Net. size (HF-C) .07 .03 .00 .02 .16 .15
8. Net. size (HF-P) .09 .11 .11 .08 .16 .02 .07
9. Net. size (NF-C) .03 .06 .01 .05 .19 .06 .03 .10
10. Net. size (NF-P) .19 .09 .22 .07 .07 .15 .11 .13 .04
11. Pos. exch. (VF-C) .18 .22 .21 .14 .32 .22 .14 .18 .25 .10
12. Pos. exch. (VF-P) .06 .02 .07 .07 .05 .63 .15 .07 .08 .14 .28
13. Pos. exch. (HF-C) .01 .01 .01 .02 .22 .18 .51 .03 .03 .01 .22 .17
14. Pos. exch. (HF-P) .14 .14 .11 .09 .16 .03 .11 .53 .09 .08 .28 .10 .06
15. Pos. exch. (NF-C) .03 .11 .01 .04 .23 .07 .02 .14 .62 .02 .28 .10 .03 .13
16. Pos. exch. (NF-P) .07 .07 .14 .01 .06 .09 .07 .02 .06 .44 .13 .05 .06 .12 .03
17. Neg. exch. (VF-C) .12 .16 .12 .12 .02 .15 .10 .06 .09 .02 .21 .15 .15 .03 .10 .04
18. Neg. exch. (VF-P) .07 .11 .05 .12 .01 .11 .06 .01 .03 .02 .07 .22 .05 .04 .04 .01 .22
19. Neg. exch. (HF-C) .12 .12 .10 .08 .12 .05 .13 .05 .03 .01 .08 .05 .22 .04 .03 .02 .04 .02
20. Neg. exch. (HF-P) .01 .05 .02 .01 .04 .00 .04 .02 .02 .03 .01 .02 .05 .09 .02 .01 .09 .02 .01
POSITIVE AND NEGATIVE EXCHANGES

21. Neg. exch. (NF-C) .01 .02 .01 .02 .03 .02 .04 .01 .08 .00 .06 .02 .02 .01 .23 .00 .02 .01 .00 .01
22. Neg. exch. (NF-P) .08 .05 .05 .05 .02 .02 .00 .02 .04 .01 .02 .03 .04 .01 .04 .10 .11 .01 .26 .00 .01
M 0.85 4.54 3.10 1.78 3.07 0.89 1.22 2.00 0.33 3.63 2.00 1.21 1.40 1.41 1.19 1.81 1.14 1.02 1.02 1.01 1.01 1.01
SD 1.26 1.18 0.94 0.80 3.76 2.30 2.60 3.37 1.14 4.00 0.86 0.47 0.64 0.56 0.53 0.65 0.32 0.17 0.12 0.13 0.13 0.09
Notes: C = close; exch. = exchanges; HF = horizontal family; neg. = negative; NF = nonfamily; Net. = network; P = peripheral; pos. = positive; VF = vertical family. Correlations |.07|, |.08|, and |.11| were significant
at alpha levels of .05, .01, and .001 respectively.
713
714 CHENG ET AL.

Table 2. Regression of Sociodemographics, Health, Network Sizes, Table 3. Regression of Sociodemographics, Health, Network Sizes,
and Social Exchanges (broken down by relationship category and and Social Exchanges (broken down by relationship category and
closeness) on Well-being Measures (N = 1,005) closeness) on Well-being Measures for Participants Without Vertical
Family Members (n = 295)
Depression Life Positive Negative
satisfaction affect affect Depression Life Positive Negative
satisfaction affect affect
Age .06 .21*** .07* .14***
Women .04 .14*** .08** .00 Age .12* .22*** .02 .14*
Married .04 .02 .02 .00 Women .03 .16** .09 .01
Socioeconomic status .01 .05 .02 .02 Married .01 .00 .03 .02
Living alone .00 .05 .03 .03 Socioeconomic status .05 .05 .08 .01
Chronic illnesses .11*** .09** .17*** .11*** Living alone .03 .08 .02 .05
IADL .14*** .07* .07* .10** Chronic .06 .14** .13** .05
Network sizes IADL .17** .04 .04 .09
VF-C .03 .08** .12*** .00 Network sizes
VF-P .04 .07* .03 .06* H, C .05 .04 .05 .06
HF-C .07** .04 .00 .04 H, P .01 .07 .10 .07
HF-P .03 .05 .09** .05 N, C .05 .04 .02 .06
NF-C .03 .02 .03 .05 N, P .23*** .10 .30*** .14**
NF-P .21*** .09** .22*** .11*** Positive exchanges
Positive exchanges H, C .13* .10* .11* .11*
VF-C .15*** .22*** .20*** .14*** H, P .08 .06 .05 .06
VF-P .05 .02 .08* .04 N, C .07 .07 .08 .06
HF-C .05 .08* .10** .10** N, P .03 .05 .02 .01
HF-P .08* .04 .00 .02 Negative exchanges
NF-C .05 .06 .05 .02 H, C .10 .13* .12* .10*
NF-P .06 .01 .02 .04 H, P .00 .02 .06** .02
Negative exchanges N, C .01 .05 .01 .00
VF-C .13** .15*** .13*** .11*** N, P .10* .01 .07 .03
VF-P .05 .07*** .03 .10** R2 .13 .19 .17 .08
HF-C .09** .11*** .07** .06*
Notes: Chronic = number of chronic conditions; C = close relationships;
HF-P .01 .02 .01 .01
IADL = instrumental activities of daily living; H = horizontal family; N = non-
NF-C .00 .04 .00 .02
family; NE = negative exchanges; P = peripheral relationships; PE = positive
NF-P .05 .00 .01 .02
exchanges. Standardized regression coefficients of the predictors are shown.
R2 .17 .22 .19 .12
*p < .05; **p < .01; ***p < .001.
Notes: C = close; HF = horizontal family; IADL = instrumental activities of
daily living; NE = negative exchanges; NF = nonfamily; P = peripheral; PE =
positive exchanges; VF = vertical family. Standardized regression coefficients
combinations of the received instrumental support items in
of the predictors are shown. order to create measures of positive exchanges that were
*p < .05; **p < .01; ***p < .001. comparable to that of negative exchanges. In none of these
analyses were the overall effects between positive and nega-
tive exchanges different.
horizontal family members. In contrast, negative exchanges
with peripheral nonfamily members were associated with
depression, but there were no other effects related to nonfa- Discussion
mily members. Thus, distant kin, who were more likely to The present findings suggested that relationship type and
be placed in the inner circle, were more important than nonkin closeness need to be considered together when trying to
to those without immediate kin. make sense of the relationship between social exchanges
and well-being. Other than nonkin, close partners tended to
have more effects on well-being than peripheral partners,
Equivalence Between Positive and Negative Exchanges: and social exchanges with close vertical family members
Overall Effects were most strongly associated with well-being. In addition,
Results of the model comparisons revealed that the over- partially supporting the domain-specific effect, but not the
all effects of positive exchanges on well-being were not dif- negativity effect, model (Ingersoll-Dayton et al., 1997), the
ferent from those of negative exchanges (for which the overall effect of positive exchanges on PA was greater than
signs of the coefficients were reversed before summation) that of negative exchanges. However, similar patterns were
across all measures of well-being except PA, summed bs = not found for other measures of well-being. For this reason,
.46 vs. .22 for positive and negative (signs reversed) ex- subsequent discussion will generally not make distinction
changes, respectively; F(1, 979) = 4.52, p < .05. However, between the two kinds of exchanges.
the asymmetry of the two measures might introduce bias Consistent with our expectations, social exchanges with
into the analysis, favoring positive exchanges that had more vertical and horizontal family members, but not those with
items. We selected all possible two-item combinations of the nonfamily members, were associated with well-being.
received emotional support items and all possible two-item Moreover, between vertical and horizontal family members,
 POSITIVE AND NEGATIVE EXCHANGES 715

the former were more consistently and strongly related to Does it mean that friends, for the most part, do not matter
well-being than the latter. On the whole, the effects of social to Chinese older adults? This is apparently not the case.
exchanges with horizontal family members were between Although the quality of interactions with friends was unrelated
those of the other two relationship types, and the importance to well-being, the number of friends was significantly
of the horizontal family became more pronounced when related to all well-being measures. Although this relation-
vertical family members were not present. Thus, for Chinese ship was specific to friends in the middle and outer circles,
older adults, having good relationships, in terms of the quality because there were so few friends placed in the inner circle,
of social exchanges, with family members is much more the effect could be said to apply to friends in general. A
important than the same with friends. Unlike friendships, larger circle of friends is important to provide socializing
blood ties cannot be discontinued easily, and this is espe- functions, and social activities have been found to be a
cially true for Chinese and Asian societies. Hence, the qual- robust predictor of well-being (Okun, Stock, Haring, &
ity of interactions with family members, whether positive Witter, 1984). Because social activities are usually emotion-
or negative, is crucial to the daily well-being of Chinese ally arousing (Larson et al., 1986), it was not surprising that
older adults. Moreover, under the context of familism, the effects of the friendship network were stronger for mea-
Chinese family members tend to live close to each other, sures of affect and depression than for life satisfaction (see
making them more physically available for mutual support Table 2). Thus, the importance of friendship may depend on
(Bian, Logan, & Bian, 1998). Thus, it should not be surpris- the specific research focus and operationalization of friend-
ing that relationships with kin carry more weight on ship support. It appears that universally, researchers who
well-being than those with nonkin. On the whole, the verti- focus on activities will likely find friendship to be quite
cal family, the horizontal family, and the nonfamily repre- rewarding (see, e.g., Larson et al., 1986; Wood & Robertson,
sent a hierarchy of relationship preference for Chinese 1978 for studies on Western samples), and the number of
older adults. Although more research in other cultures is friends may just be a proxy for the range and frequency of
needed, we expect that similar findings will be observed in social activities. Those who study the quality of social ex-
cultural groups, such as Hispanic, African, and African- changes or support functions, however, may not find robust
American, which similarly value the family and extended and consistent effects across cultures. In general, studies of
kinship (Cheng & Siankam, 2009; de Vos, 1990; Dilworth- Asian older populations have found that friends and neigh-
Anderson, 1992). bors are not considered to be reliable or effective sources of
support when needs arise (Koyano, Hashimoto, Fukawa,
Shibata, & Atsuaki, 1994; Yeung & Fung, 2007).
The Value of Friendship in Later Life
According to the present findings, the quality of interac-
tions with friends does not contribute to well-being beyond Close Versus Peripheral Partners
that with family members. The only exception was found in This is the first study examining the effect of social ex-
depression but only when vertical family members were not changes by degree of closeness. As expected, positive and
in the network. These findings echo the observation even negative exchanges with close relationship partners, those
among Western scholars about the declining significance of most central to ones life, were more impactful than periph-
friends in midlife due to responsibilities and intimacies eral partners on ones well-being. However, closeness per
within the family and perhaps also in old age when physical se was not sufficient to differentiate the importance of dif-
and financial constraints may limit interactions outside the ferent relationship partners. One must take into account the
family (Blieszner & Roberto, 2004; Carbery & Buhrmester, nature of the relationship. The order of influence, from
1998). Within Chinese (and generally Asian) cultures, strongest to weakest, was close vertical family, close hori-
Cheng and colleagues (2009) further suggested that interac- zontal family, peripheral vertical family, peripheral horizon-
tions with friends might be less well-being enhancing than tal family, and nonfamily (regardless of closeness). Thus the
those in Western contexts because of the norm of reciproc- idea of close versus less close appears to be a more impor-
ity and social harmony in these cultures. These collectivistic tant distinction for family members than it is for friends.
norms regulate all social behaviors but are more applicable Across cultures (e.g., Antonucci & Akiyama, 2004), imme-
to nonkin than to kin. Under these norms, one is discour- diate kin tend to be placed in the inner circle. Neyer and
aged from expressing ones personal concerns that may sig- Lang (2003) believe that this is due to the evolutionary
nify problems in the group or may disrupt harmony and to adaptiveness of investment in kin that enhances continua-
reciprocate when favors, whether wanted or desirable, are tion of the family line. At the same time, when an immedi-
received. The costs of seeking help from friends are there- ate kin member is excluded from the inner circle, it might
fore higher in Chinese and Asian societies than in Western signal a certain degree of detachment from the relationship.
societies. Thus, on the whole, the quality of friendship may The opposite may be true for distant kin. Distant kin tend to
not have any significant impact on the well-being of Chinese be placed in the middle or even outer circle due to decreas-
older persons. ing genetic relatedness (Neyer & Lang, 2003) or functional
716 CHENG ET AL.

significance (Cheng et al., 2009), but the most intimate ones 10 Lo Ping Road, Tai Po, New Territories, Hong Kong, China. E-mail:
[email protected].
are nonetheless placed in the inner circle. Thus, network
circle placement might carry different meanings that are be-
References
yond subjective closeness for different blood ties, a ques-
Adams, R. G., & Blieszner, R. (1995). Aging well with friends and family.
tion that has yet to be explored in the literature. Because American Behavioral Scientist, 39, 209224. doi:10.1177/000276
there were so few friends in the inner circle, there might not 4295039002008
be sufficient statistical power to render a firm conclusion on Akiyama, H., Antonucci, T., Takahashi, K., & Langfahl, E. S. (2003).
the importance of subjective closeness with friends. On the Negative interactions in close relationships across the life span. The
Journals of Gerontology, Series B: Psychological Sciences and
whole, on top of the hierarchy of relationship preference,
Social Sciences, 58, 7079.
the influences of vertical and horizontal family members Albertini, M., & Kohli, M. (2009). What childless older people give: Is the
were moderated by subjective closeness. Thus, a strong generational link broken? Ageing & Society, 29, 12611274. doi:
safety net for Chinese older adults may not be formed ex- 10.1017/S0144686X0999033X
clusively by immediate family members but by a broader Antonucci, T., & Akiyama, H. (2004). Attachment and close relationships
network of kin, including distant and immediate family, across the life span. Attachment & Human Development, 6, 353370.
doi:10.1080/1461673042000303136
who are emotionally close. Bardis, P. D. (1959). A familism scale. Marriage and Family Living, 21,
340341. doi:10.2307/347107
Bian, F., Logan, J. R., & Bian, Y. (1998). Intergenerational relations in
Study Limitations and Future Research Directions urban China: Proximity, contact, and help to parents. Demography,
The present studies suffered from several limitations. 35, 115124. doi:10.2307/3004031
First, we did not distinguish between network members Birditt, K. S., Jackey, L. M. H., & Antonucci, T. C. (2009). Longitudinal
placed in the middle circle versus those in the outer circle. patterns of negative relationship quality across adulthood. The Journals
However, had we done so, missing values would dominate of Gerontology, Series B: Psychological Sciences and Social Sciences,
64, 5564. doi:10.1093/geronb/gbn031
a number of relationship categories (e.g., vertical family, Blieszner, R., & Roberto, K. A. (2004). Friendship across the life span:
outer circle) in ways that create more intractable problems Reciprocity in individual and relationship development. In F. R. Lang &
for data analysis. In the present analyses, the low number of K. L. Fingerman (Eds.), Growing together: Personal relationships
nonfamily members in the inner circle was a case in point, across the lifespan (pp. 159182). New York: Cambridge University
making it difficult to yield reliable estimates of the effects Press.
Cantor, M. H. (1979). Neighbors and friends: An overlooked resource in the
for this category of network members. Second, future informal support system. Research on Aging, 1, 434463. doi:10.1177/
research should adopt longitudinal designs to investigate 016402757914002
possible causal relations between social exchanges and sub- Carbery, J., & Buhrmester, D. (1998). Friendship and need fulfillment during
jective well-being, which is not possible in the present three phases of young adulthood. Journal of Social and Personal
cross-sectional design. Relationships, 15, 393409. doi:10.1177/0265407598153005
Census and Statistics Department. (2007). Hong Kong 2006 population
Lastly, the centrality of kinship needs to be further inves-
by-census main report (Vols. I & II). Hong Kong, China: Author.
tigated in populations with significant frailties. As an older Chatterjee, S., & Hadi, A. S. (2006). Multiple linear regression. In
person becomes more physically dependent, family mem- S. Chatterjee & A. S. Hadi (Eds.), Regression analysis by example
bers who are more suited to provide day-to-day assistance (4th ed., pp. 5384). Hoboken, NJ: Wiley-Interscience. doi:10.1002/
(Felton & Berry, 1992) may become even more important, 0470055464.ch3
Cheng, S.-T. (2004). Age and subjective well-being revisited: A discrep-
and the roles of friends may be further diminished. For
ancy perspective. Psychology and Aging, 19, 409415. doi:10.1037/
those without immediate kin, the supportiveness of distant 0882-7974.19.3.409
kin under such circumstances remains unclear. These are Cheng, S.-T. (2009). The social networks of nursing-home residents in
important questions, but they apply only to community- Hong Kong. Ageing & Society, 29, 163178. doi:10.1017/S0144686X
dwelling persons. Once a person moves to an institution, the 0800771X
Cheng, S.-T., & Chan, A. C. M. (2004). A brief version of the Geriatric
roles of network members, and even relationships with
Depression Scale for the Chinese. Psychological Assessment, 16,
them, are redefined (for a discussion, see Cheng, 2009; 182186. doi:10.1037/1040-3590.16.2.182
Cheng, Lee, & Chow, 2010). The desirability of different Cheng, S.-T., & Chan, A. C. M. (2006). Filial piety and psychological
social ties may change as a result of changing health and well-being in well older Chinese. The Journals of Gerontology, Series
environmental contexts. B: Psychological Sciences and Social Sciences, 61, 262269.
Cheng, S.-T., Lee, C. K. L., Chan, A. C. M., Leung, E. M. F., & Lee, J.-J.
(2009). Social network types and subjective well-being in Chinese older
Funding
adults. The Journals of Gerontology, Series B: Psychological Sciences
The preparation of this manuscript was supported in part by Competitive and Social Sciences, 64, 713722. doi:10.1093/geronb/gbp075
Earmarked Research Grant No. CityU1495/05H of the Research Grants Cheng, S.-T., Lee, C. K. L., & Chow, P. K. (2010). Social support and
Council of Hong Kong.
psychological well-being of nursing home residents in Hong Kong.
International Psychogeriatrics, 22, 11851190. doi:10.1017/
Correspondence S1041610210000220
Correspondence should be addressed to Sheung-Tak Cheng, PhD, Cheng, S.-T., & Siankam, B. (2009). The impacts of the HIV/AIDS
Department of Psychological Studies, Hong Kong Institute of Education, pandemic and socioeconomic development on the living arrangements
 POSITIVE AND NEGATIVE EXCHANGES 717

of older persons in sub-Saharan Africa: A country-level analysis. exchanges: A longitudinal model with comparable measures.
American Journal of Community Psychology, 44, 136147. doi: Psychology and Aging, 18, 746754. doi:10.1037/0882-7974.
10.1007/s10464-009-9243-y 18.4.746
Cheng, S.-T., & Yim, Y.-K. (2008). Age differences in forgiveness: The role Newsom, J. T., Rook, K. S., Nishishiba, M., Sorkin, D. H., & Mahan, T. L.
of future time perspective. Psychology and Aging, 23, 676680. doi: (2005). Understanding the relative importance of positive and nega-
10.1037/0882-7974.23.3.676 tive social exchanges: Examining specific domains and appraisals.
Cheng, S.-T., Yu, E. C. S., Lee, S. Y., Wong, J. Y. H., Lau, K. H., Chan, L. K., The Journals of Gerontology, Series B: Psychological Sciences and
. . . Wong, M. W. L. (2010). The Geriatric Depression Scale as a screen- Social Sciences, 60, 304312.
ing tool for depression and suicide ideation: A replication and exten- Neyer, F. J., & Lang, F. R. (2003). Blood is thicker than water: Kinship
sion. American Journal of Geriatric Psychiatry, 18, 256265. doi: orientation across adulthood. Journal of Personality and Social
10.1097/JGP.0b013e3181bf9edd Psychology, 84, 310321. doi:10.1037/0022-3514.84.2.310
de Vos, S. (1990). Extended family living among older people in six Okun, M. A., & Keith, V. M. (1998). Effects of positive and negative social
Latin American countries. The Journals of Gerontology, Series B: exchanges with various sources on depressive symptoms in younger
Psychological Sciences and Social Sciences, 45, 8794. and older adults. The Journals of Gerontology, Series B: Psychological
Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The satis- Sciences and Social Sciences, 53, 420.
faction with life scale. Journal of Personality Assessment, 49, 7175. Okun, M. A., Stock, W. A., Haring, M. J., & Witter, R. A. (1984).
doi:10.1207/s15327752jpa4901_13 Social activity/subjective well-being relation: A quantitative
Dilworth-Anderson, P. (1992). Extended kin networks in Black fami- synthesis. Research on Aging, 6, 4565. doi:10.1177/01640275
lies. Generations: Journal of the American Society on Aging, 16, 84006001003
2932. Rook, K. S. (1984). The negative side of social interaction: Impact on
Emmons, R. A., & McCullough, M. E. (2004). The psychology of grati- psychological well-being. Journal of Personality and Social Psychol-
tude: Series in affective science. New York: Oxford University ogy, 46, 10971108. doi:10.1037/0022-3514.46.5.1097
Press. Rook, K. S. (2001). Emotional health and positive versus negative social
Felton, B. J., & Berry, C. A. (1992). Do the sources of the urban elderlys exchanges: A daily diary analysis. Applied Developmental Science,
social support determine its psychological consequences? Psychology 5, 8697. doi:10.1207/S1532480XADS0502_4
and Aging, 7, 8997. doi:10.1037/0882-7974.7.1.89 Rook, K. S., & Pietromonaco, P. R. (1987). Close relationships: Ties that
Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). Mini-Mental heal or ties that bind? W. H. Jones & D. Perlman (Eds.), Advances
State: A practical method for grading the cognitive state of patients in personal relationships (Vol. 1, pp. 135). Greenwich, CT: JAI
for the clinician. Journal of Psychiatric Research, 12, 189198. Press.
doi:10.1016/0022-3956(75)90026-6 Rook, K. S., & Schuster, T. L. (1996). Compensatory processes in the
Ingersoll-Dayton, B., Morgan, D., & Antonucci, T. (1997). The effects of social networks of older adults. In G. R. Pierce, B. R. Sarason &
positive and negative social exchanges on aging adults. The Journals I. G. Sarason (Eds.), Handbook of social support and the family
of Gerontology, Series B: Psychological Sciences and Social Sciences, (pp. 219248). New York: Plenum.
52, 190199. Skowronski, J. J., & Carlston, D. E. (1989). Negativity and extremity
Kahn, R. L., & Antonucci, T. C. (1980). Convoys over the life course: biases in impression formation: A review of explanations. Psycho-
Attachment, roles, and social support. In P. B. Baltes & O. Brim logical Bulletin, 105, 131142. doi:10.1037/0033-2909.105.1.131
(Eds.), Life-span development and behavior (Vol. 3, pp. 253286). StataCorp. (2009). Stata statistical software: Release 11. College Station,
San Diego, CA: Academic Press. TX: StataCorp LP.
Kesebir, P., & Diener, E. (2010). In pursuit of happiness: Empirical Taylor, S. E. (1991). Asymmetrical effects of positive and negative events:
answers to philosophical questions. Perspectives in Psychological The mobilization-minimization hypothesis. Psychological Bulletin,
Science, 3, 117125. doi:10.1111/j.1745-6916.2008.00069.x 110, 6785. doi:10.1037/0033-2909.110.1.67
Koyano, W., Hashimoto, M., Fukawa, T., Shibata, H., & Atsuaki, G. Thompson, M. G., & Heller, K. (1990). Facets of support related to well-
(1994). The social support system of the Japanese elderly. Journal of being: Quantitative social isolation and perceived family support in a
Cross-Cultural Gerontology, 9, 323333. doi:10.1007/BF00978217 sample of elderly women. Psychology and Aging, 5, 535544.
Krause, N., & Rook, K. S. (2003). Negative interaction in later life: Issues doi:10.1037/0882-7974.5.4.535
in the stability and generalizability of conflict across relationships. Tse, M. C., & Cheng, S.-T. (2006). Depression reduces forgiveness
The Journals of Gerontology, Series B: Psychological Sciences and selectively as a function of relationship closeness and transgression.
Social Sciences, 58, 8899. Personality and Individual Differences, 40, 11331141. doi:10.1016/
Lang, F. R., & Carstensen, L. L. (1994). Close emotional relationships in j.paid.2005.11.008
later life: Further support for proactive aging in the social domain. Wenger, G. C. (2009). Childlessness at the end of life: Evidence from
Psychology and Aging, 9, 315324. doi:10.1037/0882-7974.9.2.315 rural Wales. Ageing & Society, 29, 12431259. doi:10.1017/
Larson, R., Mannell, R., & Zuzanek, J. (1986). Daily well-being of older S0144686X09008381
adults with friends and family. Psychology and Aging, 1, 117126. White, H. (1980). A heteroskedasticity-consistent covariance matrix
doi:10.1037/0882-7974.1.2.117 estimator and a direct test for heteroskedasticity. Econometrica, 48,
Lawton, M. P., & Brody, E. M. (1969). Assessment of older people: Self- 817838. doi:10.2307/1912934
maintaining and instrumental activities of daily living. The Geron- Wood, V., & Robertson, J. F. (1978). Friendship and kinship interaction:
tologist, 9, 179186. Differential effect on the morale of the elderly. Journal of Marriage
Li, L. W., & Liang, J. (2007). Social exchanges and subjective well-being and the Family, 40, 367375. doi:10.2307/350767
among older Chinese: Does age make a difference? Psychology and Yang, K.-S. (2006). Indigenized conceptual and empirical analyses of
Aging, 22, 386391. doi:10.1037/0882-7974.22.2.386 selected Chinese psychological characteristics. International Journal
Merz, E.-M., & Huxhold, O. (2010). Wellbeing depends on social relation- of Psychology. Special Issue: The Indigenous Psychologies, 41,
ship characteristics: Comparing different types and providers of sup- 298303.
port to older adults. Ageing & Society, 30, 843857. doi:10.1017/ Yeung, G. T. Y., & Fung, H. H. (2007). Social support and life satisfac-
S0144686X10000061 tion among Hong Kong Chinese older adults: Family first?
Newsom, J. T., Nishishiba, M., Morgan, D. L., & Rook, K. S. (2003). The European Journal of Ageing, 4, 219227. doi:10.1007/s10433-
relative importance of three domains of positive and negative social 007-0065-1
718 CHENG ET AL.

Appendix model. The F value and the degrees of freedom for testing
The F test for model comparison can be expressed in the hypothesis are
terms of the variances explained and the degrees of freedom R p2 Rq2
of the full and constrained models, where R p2 is the variance
p q
explained in the full model and Rq2 is the variance explained F= , df = p q n p 1
in the constrained model, p is the degrees of freedom of the 1 R p2
full model and q is the degrees of freedom of the constrained n p 1