4. Results
The descriptive statistics of the sociodemographic covariates of students’ gender, family composition, and ethnic origins, and the study variables of students’ educational expectations, science learning performance, and the successful completion of a STEM degree are shown in
Table 1. Females and males shared 48% (
n = 1495) and 52% (
n = 1621), respectively, and 87.1% of the students came from two-parent families (
n = 2715) compared to 12.9% of their peer classmates from other family structures (
n = 401). For ethnic origins, whites were the majority (
n = 2169, 69.6%), followed by African Americans (
n = 504, 16.2%), Hispanics (
n = 284, 9.1%), Asians (
n = 112, 3.5%), and Native Americans (
n = 47, 1.5%). The mean scores of students’ educational expectations present a slightly descending pattern with an up-turn in the least year of secondary school:
= 4.005 in grade 8, and
= 3.886 in grade 9,
= 3.717 in grade 10,
= 3.612 in grade 11, and
= 3.731 in grade 12. In addition, the average grade levels of students’ science learning performance across the years of secondary school also display a similar changing pattern to that of educational expectations with an up-turn in grade 11 and grade 12:
= 5.997 in grade 8,
= 5.843 in grade 9, and
= 5.629 in grade 10, 5.672 in grade 11, and 5.840 in grade 12. Moreover, only 7.7% of the students (
n = 239) obtained a four-year STEM baccalaureate degree in 2009.
Before conducting PP-LGCM modeling, correlations of the study variables of students’ educational expectations and science learning performance in secondary school and their successful completion of a STEM degree in adulthood were run.
Table 2 shows that the same- and cross-domain correlations of the study variables were all significant at
p < 0.001. The same-domain correlations of students’ educational expectations from grade 8 to grade 12 ranged from r = 0.595 to 0.834 with a decreasing pattern of intrapersonal stability observed, in which the one-year lagged correlations of students’ educational expectations were from r = 0.731 to 0.824, and the two-year and three-year lagged correlations were from r = 0.689 to 0.747, and r = 0.640 to 0.680, respectively. In addition, the same-domain correlations of students’ science learning performance from grade 8 to grade 12 ranged from r = 0.311 to 0.592, which also show the same decreasing pattern of intrapersonal stability as to the same-domain correlations of students’ educational expectations, in which the one-year lagged correlations of students’ science learning performance were from r = 0.461 to 0.614, and the two-year and three-year lagged correlations were from r = 0.402 to 0.501, and r = 0.343 to 0.409, respectively. In fact, the decreasing levels of the lagged correlations among the same-domain measures across time verify the validity of their variability [
75,
76]. Besides, the cross-domain correlations between students’ educational expectations and science learning performance across the years of secondary school ranged from r = 0.323 to 0.394, which are smaller than those of the same-domain correlations but had a consistent decreasing pattern of correlations akin to the same-domain correlations. The concurrent cross-domain correlations of students’ educational expectations with science learning performance ranged from r = 0.291 to 0.351, and the one-year, two-year, three-year, and four-year lagged correlations ranged from r = 0.295 to 0.359, r = 0.242 to 0.310, r = 0.228 to 0.268, and r = 0.204 and 0.223, respectively. For the correlations between students’ successful completion of a STEM degree in adulthood and their educational expectations and science learning performance across the years of secondary school, the point-biserial correlation coefficients ranged from r = 0.184 to 0.243.
Table 3 shows the results of the univariate direct PP-LGCM model investigating the effects of students’ educational expectations on students’ science learning performance in secondary school (PP-LGCM Model 1A), for which a very good model fit was obtained: CFI = 0.972; RMSEA = 0.062; SRMR = 0.048. Nevertheless, the modification indices suggested setting the covariances between the residuals of students’ grade 9 and grade 10 educational expectations, grade 9 and grade 10 science learning performance, and grade 10 and grade 11 science learning performance, in which an excellent model fit appeared: CFI = 0.984; RMSEA = 0.048; SRMR = 0.039. The intercept factor loadings of students’ educational expectations across the years of secondary school from grade 8 to grade 12 were λ = 0.868, 0.866, 0.888, 0.901, and 0.862,
p < 0.001; and the slope factor loadings from grade 9 to grade 12 were λ = 0.151, 0.310, 0.473, and 0.603,
p < 0.001. The intercept factor loadings of students’ science learning performance across the years of secondary school from grade 8 to grade 12 were λ = 0.714, 0.728, 0.727, 0.761, and 0.824,
p < 0.001; and its slope factor loadings from grade 9 to grade 12 were λ= 0.177, 0.354, 0.556, and 0.803,
p < 0.001. For the within-domain regression effects, the slope factors of students’ educational expectations and science learning performance were significantly predicted by their intercept factors, β = −0.247 and −0.451,
p < 0.001, suggesting that the growth trajectories of students’ educational expectations and science learning performance during the years of secondary school were susceptible to the impacts of students’ initial development in educational expectations and science learning performance in secondary school. This means that students with a higher initial development of educational expectations and science learning performance appeared to have a slower rate of growth in educational expectations and science learning performance across the years of secondary school later. For the cross-domain regression effects, the intercept factor of students’ educational expectations was significantly and substantially predictive of the intercept factor of students’ science learning performance in a positive way, β = 0.633,
p < 0.001, indicating that students with a higher developmental trajectory of educational expectations contributed to their better developmental trajectory of science learning performance in secondary school. Furthermore, the slope factor of students’ educational expectations significantly and positively predicted the slope factor of students’ science learning performance, β = 0.216,
p < 0.001, meaning that students with a better growth trajectory of educational expectations contributed to their better growth in science learning performance across the years of secondary school.
Next, the conditional direct PP-LGCM model was examined to predict the distal outcome of students’ successful completion of a STEM degree in adulthood by the developmental and growth trajectories of students’ educational expectations and science learning performance in secondary school while simultaneously adjusting for the sociodemographic covariates of students’ gender, family composition, and ethnic origins (PP-LGCM Model 1B).
Figure 1 shows the results of PP-LGCM Model 1B, in which an excellent model fit was obtained: CFI = 0.988; RMSEA = 0.031; SRMR = 0.024. Specifically, the within- and cross-domain regression effects for the relationships between the developmental and growth trajectories of students’ educational expectations and science learning performance, plus their intercept and slope factor loadings, were similar to those of PP-LGCM Model 1A. Notably, the intercept factors of students’ educational expectations and science learning performance were both significantly and positively predictive of students’ successful completion of a STEM degree in adulthood, β = 0.145 and 0.244,
p < 0.001, indicating that a unit increase in the developmental trajectories of students’ educational expectations and science learning performance resulted in the higher odds of students’ later graduation with a STEM degree by 15.6% and 27.6%, respectively. Furthermore, the slope factors of students’ educational expectations and science learning performance were also significantly and positively predictive of students’ successful completion of a STEM degree in adulthood, β = 0.083 and 0.152,
p < 0.001, meaning that a unit increase in the growth trajectories of students’ educational expectations and science learning performance contributed to their higher odds of graduation with a STEM degree in adulthood by 8.6% and 16.4%, respectively. For the effects of students’ sociodemographic covariates, male students, compared to their female classmates, significantly had the higher developmental trajectory of science learning performance, β = 0.080,
p < 0.01, and were also more likely to graduate with a STEM degree, β = 0.077,
p < 0.01, OR = 1.080 (refer to
Table A1 in the
Appendix A). Moreover, students living in a two-parent family tended to have a higher growth trajectory of educational expectations and a higher developmental trajectory of science learning performance, β = 0.054 and 0.053,
p < 0.05, and were also more likely to graduate with a STEM degree, β = 0.032,
p < 0.05, OR = 1.032. In addition, compared to their Asian classmates, Hispanic, African American, and Native American students had a significantly lower developmental trajectory of educational expectations, β = −0.204, −0.104, −0.165, and −0.084,
p < 0.001, 0.001, 0.01, and 0.01, and science learning performance in secondary school, β = −0.113, −0.161, −0.109, and −0.075,
p < 0.001, 0.001, 0.01, and 0.01. Furthermore, African American and Native American students, in contrast to their Asian counterparts, had a significantly lower growth trajectory of science learning performance during the years of secondary school, β = −0.157 and −0.059, p < 0.05.
On the other hand, the univariate inverse PP-LGCM model was conducted to test the effects of students’ developmental and growth trajectories of science learning performance on their developmental and growth trajectories of educational expectations in secondary school (PP-LGCM Model 2A).
Table 4 shows that an excellent model fit was obtained: CFI = 0.974; RMSEA = 0.059; and SRMR = 0.044. Nevertheless, the modification indices recommended setting covariances between the residuals of students’ grade 9 and grade 10, as well as grade 10 and grade 11 science learning performance, for which a better model fit was obtained: CFI = 0.987; RMSEA = 0.043; and SRMR = 0.033. The intercept factor loadings of students’ science learning performance in secondary school from grade 8 to grade 12 were λ = 0.715, 0.727, 0.726, 0.759, and 0.823,
p < 0.001, and the intercept factor loadings of students’ science learning performance from grade 8 to grade 12 were λ = 0.871, 0.891, 0.913, 0.918, and 0.879,
p < 0.001. In addition, the slope factor loadings of students’ science learning performance across the years of secondary school from grade 9 to grade 12 were λ = 0.175, 0.349, 0.547, and 0.790,
p < 0.001, and the slope factor loadings of students’ educational expectations were λ = 0.159, 0.326, 0.491, and 0.627,
p < 0.001. For the within-domain regression effects, the intercept factors of students’ science learning performance and educational expectations were significantly predictive of its slope factors, respectively, β = −0.445 and −0.463,
p < 0.001, which means that the growth trajectories of students’ science learning performance and educational expectations were influenced by their developmental trajectories over the years of secondary school. Specifically, students with a higher initial development of science learning performance and educational expectations in secondary school presented a slower rate of growth in science learning performance and educational expectations afterward. For the cross-domain regression effects, the intercept factor of students’ science learning performance was significantly and positively predictive of both the students’ intercept and slope factors of educational expectations, β = 0.589 and 0.370,
p < 0.001, which means that students of higher initial development of science learning performance contributed to their better developmental and growth trajectories of science learning performance in the years of secondary school. In addition, the slope factor of students’ science learning performance was significantly and positively predictive of the students’ slope factor of educational expectations, β = 0.370,
p < 0.001, which indicates that students of progressive improvement in science learning performance over time also promoted their continuing enhancement of educational expectations across the years of secondary school.
Furthermore, the conditional inverse PP-LGCM model was examined to predict the distal outcome of students’ successful completion of a STEM degree in adulthood (PP-LGCM Model 2B). An excellent model fit was obtained: CFI = 0.988; RMSEA = 0.030; SRMR = 0.023 (
Figure 2). The within-domain regression effects and intercept factor loadings, as well as the slope factor loadings of students’ science learning performance and educational expectations, were similar to those of PP-LGCM Model 2A. For the cross-domain regression effects, the intercept factor of students’ science learning performance significantly and positively predicted students’ intercept and slope factors of educational expectations, β = 0.610 and 0.367,
p < 0.001. In addition, the intercept factor of students’ science learning performance and the intercept and slope factors of students’ educational expectations were jointly and significantly predictive of students’ successful completion of a STEM degree in adulthood, β = 0.267, 0.129, and 0.074,
p < 0.001, 0.001, and 0.01. This means that a unit increase in the developmental trajectory of students’ science learning performance and the developmental and growth trajectories of students’ educational expectations contributed to the higher odds of students’ later graduation with a STEM degree in adulthood by 30.6%, 13.7%, and 7.6%, respectively. Furthermore, the slope factor of students’ science learning performance was also significantly and positively predictive of the slope factor of students’ educational expectations, β = 0.148,
p < 0.01, and the successful completion of a STEM degree, β = 0.161,
p < 0.001, in which the latter means that a unit increase in the growth trajectory of students’ science learning performance contributed to their higher odds of the acquisition of a STEM degree in adulthood by 17.4%. Furthermore, some of the students’ sociodemographic effects on their developmental and growth trajectories of science learning performance and educational expectations, as well as the successful completion of a STEM degree, remain similar to those of PP-LGCM Model 1B (refer to
Table A2 in the
Appendix A for the details). Specifically, male students, compared to their female classmates, had a significantly higher developmental trajectory of science learning performance, β = 0.017,
p < 0.1, and were also more likely to graduate with a STEM degree, β = 0.080,
p < 0.05, OR = 1.080. Furthermore, students in two-parent families had a higher growth trajectory in educational expectations, β = 0.095,
p < 0.01. Moreover, compared to their Asian classmates, Hispanic, African American, and Native American students had a significantly lower developmental trajectory of science learning performance, β = −0.283, −0.332, −0.225, and −0.091,
p < 0.001, 0.001, 0.001, and 0.05.
5. Discussion
The current study is the first of its kind to investigate the reciprocal influences of students’ developmental and growth trajectories of educational expectations and science learning performance directly and inversely on each other during the years of secondary school and how they contribute to the students’ distal outcome of successful completion of a STEM degree over time in mid-adulthood. Due to the malleability and cultivability of educational expectations and science learning performance among students in the course of secondary school, appropriate educational designs and policy interventions should be implemented and promoted to help secondary school students develop and establish better educational expectations and academic performance in different science subjects [
14,
43,
44]. Generally, existing cross-sectional and short-term longitudinal research tends to investigate the relationships between educational expectations and the academic development of students by using correlational or cross-lagged effect designs to treat their effects as fixed and stable [
2,
14,
39,
77]. These studies manifestly overlooked the developing and changing nature of students’ educational expectations and academic development over time. In addition, little investigation has examined the relationships between students’ educational expectations and academic development, e.g., science learning performance, over the years of secondary school in relation to their later educational achievement in mid-adulthood, e.g., successful graduation with a STEM degree. Furthermore, as science education and the promotion of students’ STEM development are of interest to educators and policymakers because of their profound implications for the cultivation of STEM intellectuals to promote societal, economic, cultural, and technological advancements [
12,
28,
55], more research is needed to scrutinize how the developmental and growth trajectories of students’ educational expectations and science learning performance during the years of secondary school may affect students’ later STEM success in adulthood. In the current study, both the conditional direct and inverse PP-LGCM models showed that the educational expectations and science learning performance of students impacted each other reciprocally in a direct and inverse way, which then jointly predicted students’ successful completion of a STEM degree 15 years later, in mid-adulthood. This reveals the importance of continuity and mutuality in the relationship between students’ academic motivation and educational development in the early years and their profound impacts carrying over on their later academic achievement extensively over time.
One of the main findings in the current study is that the developmental and growth trajectories of students’ educational expectations and science learning performance during the years of secondary school significantly and positively reinforce each other and then contribute to students’ later completion of a STEM degree in adulthood. For the developing and changing processes of students’ educational expectations and science learning performance in secondary school, the current study found that the growth trajectories of students’ educational expectations and science learning performance across the years of secondary school are susceptible to their own initial developmental trajectories in secondary school. This is congruent with a few more advanced longitudinal studies that point to the progressive nature of students’ academic motivation and educational development, formulated in the early years of secondary school [
22,
23,
48]. As such, educators and adolescent practitioners should pay attention to the promotion strategies and pro-educational environments that can assist students to cultivate and enhance their educational expectations and science learning continuously and progressively in the course of secondary school, as these are profoundly influential on their later STEM achievement in adulthood [
78,
79]. The findings of the current study are consonant with existing longitudinal research with cross-lagged designs that the developmental and changing processes of students’ academic motivation and educational performance are mutually reinforced [
26,
48,
80,
81]. Nevertheless, the superiority of the current study compared to the existing longitudinal research studies of cross-lagged designs is that it takes a longer time span to examine the development and changes of students’ educational expectations and science learning performance across the years of secondary school and treats them as the latent developmental and growth trajectories to predict students’ later STEM achievement in adulthood. Hence, the findings of the current study give insights into the need for providing students with lower development of educational expectations and science learning performance, in the initial stage of secondary school, a more responsive and supportive educational initiative to avoid their subsequently accumulated academic lag due to their detrimental impacts on students’ STEM development in adulthood. In fact, the reciprocal relationships between the developmental and growth trajectories of students’ educational expectations and science learning performance supported in the current study reveal their cultivability and malleability, which should be targeted for educational innovations and policy reforms to help establish a stimulating and positive learning context to promote students’ educational expectations and science learning performance to better prepare them for later STEM achievement.
Moreover, the reciprocal relationship between the developmental and growth trajectories of students’ educational expectations and science learning performance is corroborated in the current study, in which students’ educational expectations and science learning performance shaped each other in a positive way over the course of secondary school. This is proved in the conditional direct and inverse PP-LGCM models (PP-LGCM Model 1B and PP-LGCM Model 2B), which reveal the importance of promoting students’ educational expectations and science learning performance simultaneously due to the bidirectionality between them. Nevertheless, stronger parallel cross-domain regression effects from students’ educational expectations to science learning performance were observed in both the univariate and conditional direct PP-LGCM Models (PP-LGCM Model 1A and 1B) in contrast to the inverse PP-LGCM Models (PP-LGCM Model 2A and 2B). This means that the effects of students’ developmental and growth trajectories of educational expectations on their developmental and growth trajectories of science learning performance compared to the inverse effects of students’ developmental and growth trajectories of science learning performance on their developmental and growth trajectories of educational expectations are manifestly different, such as β = 0.622 vs. 0.615 for the developmental trajectories and β = 0.215 vs. 0.148 for the growth trajectories. This explicates the robustness of the development and growth of students’ educational expectations in contribution to the development and growth of their science learning performance across the years of secondary school. Furthermore, only the transitioning cross-domain regression effect from students’ science learning performance to their educational expectations was significant in the conditional inverse PP-LGCM Model (PP-LGCM Model 2B), which sheds light on the importance of supporting students to perform better academically in secondary school as a learning drive to boost their growth of educational expectations that, in turn, contribute to their growth of science learning performance and successful graduation with a STEM degree in adulthood.
In addition, the current study found that the developmental and growth trajectories of students’ educational expectations and science learning performance were significantly predictive of the higher odds of students’ completion of a STEM degree in adulthood, which indicates that maintaining both high educational expectations and science learning performance and improving them progressively are pivotal for students to later obtain a STEM degree. Furthermore, the effects of students’ developmental and growth trajectories of science learning performance on their successful graduation with a STEM degree are explicitly stronger than those of the students’ developmental and growth trajectories of educational expectations in both the conditional direct and inverse PP-LGCM Models (PP-LGCM Model 1B and PP-LGCM Model 2B), which reveal that keeping up science learning performance at a higher level and enhancing it persistently over the years of secondary school are crucial for students to have better STEM achievement in adulthood.
On the other hand, the effects of the sociodemographic covariates of students’ gender, family composition, and ethnic origins on their developmental and growth trajectories of education expectations and science learning performance in secondary school and the completion of a STEM degree in adulthood are noteworthy. First, male students, compared to their female classmates, had a higher developmental trajectory of science learning performance and were more likely to graduate with a STEM degree, which induces a concern for how to promote science education and STEM development among the population of female students, although we know that female students are reported to have higher general educational performance [
35,
36]. In addition, the advantages of the better development of science learning performance and the growth of educational expectations are observed for students in two-parent families compared to their classmates in other family structures. It is important for educators and adolescent practitioners to create a pro-learning environment for students from disadvantaged family backgrounds to promote their academic motivation, science learning interests, and later STEM development [
82,
83]. Moreover, Hispanic, African American, white, and Native American students, compared to their Asian classmates, are found in the current study to generally have lower developmental trajectories of educational expectations and science learning performance in the years of secondary school, which may adversely impact their later STEM achievement in adulthood. This may be due to stronger family socialization for academic outperformance in Asian culture [
84,
85]. Nevertheless, more research is needed to scrutinize these ethnic differences related to educational motivation and STEM attainments.