the self-esteem measure scores. MVQ

the self-esteem measure scores. MVQUE is a variation of the Restricted Maximum
Likelihood estimation technique and it was chosen because there is no weighting of the
random effects, so an iterative solution for estimating their variance components is not
required. Like correlation, the basic goal of variance component estimation is to assess
the population covariation between random factors and a dependent variable. The
advantage of MVQUE is that it is not limited to linear relationships between random
factors and dependent variables.
In our first analysis, the RSES scores of 26,611 individuals are treated as a
dependent variable and the participants’ gender, age group and the type of samples
(nationally representative R3+R4
vs. self-recruited Internet I5) as three random
variables. The results show that less than 11% of the total variance in the RSES scores is
attributable to the random factors (i.e., age group, gender, and sample type), only 0.41%
to the interaction between various random factors, 7.74% to differences in the type of
sample and, finally, only 3% to age differences (see Table 2). Interestingly, there are
two items (#1, “I feel I’m a person of worth, at least on an equal plane with others” and
#10, “At times I think that I am no good at all”) that demonstrate age dependence
exceeding the level of 3%. The endorsement of these two items increase with age (Table
2, last column). Moreover, these two items also demonstrate the largest effect of
sample, 10% and 8%, respectively, indicating that items that are the most sensitive to
age are simultaneously most vulnerable to the sampling bias. Five items, including both
positively and negatively worded statements, show very modest (less than 1%) total
variance caused by age.
Insert TABLE 2 about here

Finally, we analyzed similarity of the life span trajectories by a correlational
analysis. In Table 3 cross-correlations between the item trajectories of the representative
SELF-ESTEEM ACROSS THE LIFE SPAN 16

(R3+R4) and self-recruited Internet sample (I5) are shown. The Pearson correlation
evaluates only similarity in shape being insensitive to the mean level and scatter. On the
basis of this it is possible to claim that the first 5 RSES items in the Internet sample
behave differently from the respective items in the representative sample (the main
diagonal). Only trajectories of three items #6 (“I take a positive attitude toward
myself”), #9 (“I certainly feel useless at times”), and #10 (“At times I think that I am no
good at all”) are sufficiently similar in the both samples. If one observes the trajectory
of the item #6, self-esteem would appear to increase gradually from age 12 to 90
through the whole life-span. This is a replicable finding: The correlation across the nine
age groups for random vs. Internet samples was .96, p < .001. But if you happened to
choose item #9, you may think that self-esteem grows in the first half of life, and then at
age 30 starts to decline and again to increase after age 50 (except for a drop after age
70). This pattern, too, is replicable across sample types, r = .71, p = .032. However,
these two patterns of age trends are unrelated, r = .31, p = .41 and r = .39, p = .29. This
is a demonstration that single items may give replicable findings in very large samples,
but they are not necessarily generalizable to other, ostensibly similar items.
Insert TABLE 3 about here

DISCUSSION
Taken together, this research makes three essential contributions to the debate whether
there are systematic normative age differences in global self-esteem. First, we presented
some results that challenge generalization of the recently reported self-esteem trajectory
(Robins et al., 2002; Robins & Trzesniewski, 2005; Trzesniewski, Robins, Roberts, &
Caspi, 2004) across different cultures. Second, we provided evidence that the self-
SELF-ESTEEM ACROSS THE LIFE SPAN 17

recruited Internet data collection method is essentially biased compared to a random
sampling from the general population. Finally, we found that different items measuring
global self-esteem have dissimilar trajectories across the life span and therefore, the
single-item measures, like the SISE (Robins et al., 2001), may not be generalizable
across samples and comparable in similar way concerning questionnaires consisting of
several self-esteem items.
Does Global Self-esteem Have a Universal Trajectory across the Life Span?
Surprisingly, none of the four normative self-esteem trajectories in this study resembled
any other trajectory portrayed in Figure 1. Curiously, the most dissimilar were
trajectories of two large Internet samples, the multinational English-speaking Internet
sample (Robins et al., 2002) and the Estonian Internet sample (I5, this study). There are
many possible reasons for this discrepancy. The first possibility, of course, is the
difference between instruments used for measuring of global self-esteem. The
multinational study was based on a single answer to the SISE item whereas Estonian
participants responded to the RSES containing 10 items. As we demonstrated, the
normative trajectories of individual items through age were remarkably different
indicating that the wording of self-esteem items is sensitive to age. This suggests that
different age groups have a tendency to respond differently to the RSES items. For
example, older respondents agreed less frequently with the statement about feeling
useless from time to time (#9) compared to younger participants. At the same time,
some other items (e.g., #7: “On the whole, I am satisfied with myself”) did not exhibit
remarkable differences between different age groups. There have been several previous
attempts to differentiate various facets inside global self-esteem. For instance, it was
proposed that global self-esteem consists of two subfactors, positive and negative self-
SELF-ESTEEM ACROSS THE LIFE SPAN 18

esteem, while positively and negatively worded items in the RSES have a tendency to
group into two separate factors (e.g., Kohn & Schooler, 1969; Owens, 1994). However,
the division of the self-esteem trajectories into distinct types does not correspond to the
distinction between positive and negative self-esteem. Anyhow, the results of this study
support the opinion proposed by several researchers (e.g., Carmines & Zeller, 1979;
Corwyn, 2000; Dunbar, Ford, Hunt, & Der, 2000; Greenberger, Chen, Dmitrieva, &
Farruggia, 2003; Horan, DiStefano, & Motl, 2003; Marsh, 1996; Tomas & Oliver,
1999) that the wording of items has a systematic and enduring effect on the
measurement of global self-esteem.
In addition to the differences in measurement instruments (i.e., SISE vs. RSES), the
discrepancy between the results of the multinational and Estonian Internet samples
could be due to cultural and/or language differences, the pinpointing and explanation of
which is beyond the scope of this article. Furthermore, the life span trajectories of two
Estonian samples, R1* and R2*, randomly drawn from the National Census were rather
dissimilar although the identical single-item measure was used. This may mean that
there is no universal cross-sectional self-esteem trajectory through the age range and the
observed differences between different age groups remain well within the limits of the
overall measurement accuracy. Thus, the results of this study do not support the idea of
a single and invariable trajectory of global self-esteem across the human life span. If
these data support any trajectory it is flat one with some random perturbations from the
mean level.
How Typical are Self-Recruited Internet Samples?
Gosling and his colleagues (2004) argued that accumulating evidence indicates on the
consistency between the Internet and traditional pencil-and-paper methods of gathering