claim that the form of trajectories

claim that the form of trajectories is not identical since the coefficient was negative [r(8)
= -.36, p = .382].
It is noteworthy that although two Estonian samples, R1* and R2*, were both
randomly drawn from the National Census and exactly the same question was used, the
mean levels of the SISE differed slightly across the life span. The maximum difference
reached 0.51 scale among the oldest age group (age 70-79 years, d = .51, p < .001) and
ANOVA revealed that the overall difference between these two samples was significant,
F(1, 339)=13,437, p < .001. After adjustment for educational level and gender, the
ANCOVA revealed a significant effect for the samples across age groups, F(7, 2821) =
2.32, p < .05. The mean values of the SISE scale across age groups for these samples are
reported in Table 1.
Insert FIGURE 1 about here
Age Differences in Self-Esteem Measured by the Rosenberg Self-Esteem Scale
On the top in Figure 1 are presented two cross-sectional age trajectories measured by
the Estonian version of the RSES for the two types of samples: nationally representative
sample and the Estonian self-recruited Internet test-takers. The upper curve (circles)
demonstrates mean-level changes of self-esteem across the age range based on data
received from 23,248 individuals who volunteered to fill in the questionnaire at an
Internet site (I5) and the curve marked with triangles is an aggregate of two nationally
representative (R3+R4) samples.
Generally, self-esteem level of adolescents in the self-recruited Internet sample (I5)
roses gradually until it reaches a plateau in young adulthood and retaines this level (4 ±
1.5% points) till the end of the age range (correlation with age was about zero). Thus,
SELF-ESTEEM ACROSS THE LIFE SPAN 13

opposite to the multinational Internet sample (Robins et al., 2002), self-esteem is at its
lowest level during childhood among the Estonian self-recruited Internet sample.
The mean values of global self-esteem scale are systematically higher in the
Estonian self-recruited Internet sample compared to nationally representative samples.
Namely, the mean score of the RSES of the representative sample is systematically and
statistically significantly (p < .05) below the mean self-esteem level in the Estonian self-recruited Internet sample across the age groups from 13 to 59 years (Cohen's d = .18 -
.47). In total, the RSES mean in the Internet sample is statistically significantly (p <
.0001) higher compared to the nationally representative samples, d = .42 (see Table 1).
Therefore, the results of this study indicate that Estonians with a higher level of
self-esteem are more likely to answer the self-esteem questionnaire posted on the
Internet than those who are randomly or semi randomly selected from the general
population. What could be a possible explanation for this discrepancy? One possibility
is that the Estonian self-recruited Internet sample mostly consisted of individuals who
were better educated and positioned at the higher level of the social stratification than
average as it is known that such individuals have generally higher global self-esteem
(see Twenge & Campbell, 2002). As the respondents of the Internet sample did not
provide information about their educational level, this hypothesis can not be tested in
this study. However, the results of ANCOVA revealed that in the adult representative
sample (R3) more highly educated individuals tended to have a higher self-esteem after
adjustment for gender, F(2, 646) = 5.14, p < .01. More specifically, participants with a
university degree had statistically significantly (p < .0001) higher self-esteem measured
by the RSES (M = 4.02, SD = .55) compared to adults who had completed only basic or
secondary level of education (M = 3.77, SD = .67 and M = 3.86, SD = .64, respectively).
Interestingly, there was no significant difference in mean scores of the RSES between
SELF-ESTEEM ACROSS THE LIFE SPAN 14

individuals in the Internet sample and participants with a university degree in the
representative adult sample (d = .03, ns). To conclude, the results of the current study
confirm that Estonians with a higher educational level have higher self-esteem than
persons with less educational attainment. As the mean level of the RSES in the self-recruited Internet sample is comparable to those who had a university degree in the
nationally representative sample, it is very likely that the Internet sample also consisted
mostly of highly educated participants.
Item Life Span Trajectories
The mean scores of the RSES are considerably higher than the mean value of the SISE
for all of the samples. This indicates that individuals are inclined to agree with the SISE
statement “I have high self-esteem” more reluctantly than with most other 10 items
included in the Estonian version of the RSES. Indeed, a more detailed inspection shows
that the individual RSES items have not only different levels of endorsement but they
also have quite different life span trajectories: some of the items are relatively stable
with minimal variations in the mean levels from childhood to older age, the means of
other items increase across age groups and finally, the third group of items reach the
maximum value in early adulthood and start to decline afterwards. The cross-sectional
developmental trajectories across different age groups for the all 10 RSES items for a
representative (R3+R4) and self-recruited Internet sample (I5) are presented in Figure 2.
Insert FIGURE 2 about here

Next we investigated what amount of variance in respondents’ self-esteem scores
could be attributed to differences in age, sex, and sampling procedures as well as to the
interactions of these effects. To answer this question, a variance component analysis
with Minimum Variance Quadratic Unbiased Estimation (MVQUE) was performed for