Page 7 - i1052-5173-27-6

Table 2. Results of post-hoc analyses and ANCOVA

F[3, 589] = 17.55, p < 0.01, adj. R2 = 0.40

pre-intervention post-intervention
n mean* s.d.* mean s.d.

3 AR field trips (ARFTs) 218 48.8 5.3 58.1 9.3

2 ARFTs 55 48.6 4.7 55.0 11.0

1 ARFT 217 47.0 5.1 51.6 10.0

control 104 47.2 5.7 50.0 11.4

comparison of completed ARFTs contrast std. err. t P > |t|

1 vs 0 1.68 0.98 1.72 0.51

2 vs 0 3.35 1.37 2.45 0.09

3 vs 0 6.26 0.98 6.38 0.00

2 vs 1 1.66 1.24 1.34 1.00

3 vs 1 4.57 0.79 5.77 0.00

3 vs 2 2.91 1.24 2.36 0.11

ARFTs—Artificial reality field trips.

*Mean and standard deviation (s.d.) on a scale from 0–70.

Figure 2. Results of pre- and post-intervention theoretically important variables for the In an effort to determine what predicts
Geoscience Interest Survey scores for students nested regression analyses. students’ interest in the geosciences, we
having completed zero (n = 104), one (n = 217), ran a hierarchical linear model (HLM).
two (n = 55), or three (n = 218) AR field trip mod- First-order examination of the pre- and Expanding on the basic idea of regression
ules (see Table 1). post-intervention GeoIS scores shows a with a set of predictor variables and an
trend of increased student interest across outcome, HLM accounts for data that are
matrix of the variables; (2) running an all participants (Fig. 2). There is a dis- nested (Raudenbush and Bryk, 2001). In
analysis of covariance (ANCOVA) to tinctly greater increase in student interest this case, students came from different
determine the degree of impact of the AR among those participants who completed schools with different instructors and dif-
field trips on student interest; and (3) run- two and three AR field trips over those ferent regional geologic features that can
ning a hierarchical linear model (HLM) to who completed only one or were in control play a role in curriculum decisions. The
determine the predictors of student interest. groups (Fig. 2). In order to test for differ- HLM adjusted for school differences by
ences empirically, we used an Analysis of using two levels (site and student) with six
We assessed the inter-item reliability of Covariance (ANCOVA). As recommended predictors of geoscience interest: (1) GeoIS
the GeoIS by means of a Cronbach’s alpha when students are not randomly assigned pre-intervention score; (2) number of AR
analysis. While test–re-test reliability (Campbell and Stanley, 1963), we con- field trips completed; (3) site classification;
between pre- and post-tests was a possibil- trolled for preexisting differences by using (4) gender; (5) race; and (6) STEM major.
ity, we felt that inter-item reliability was the pre-test as a covariate. The results of After a null model (Table 3) that ignored
more insightful given that everyone was the ANCOVA (Table 2) indicate that the the predictors, subsequent models explored
exposed, and change was anticipated. number of field trips completed does play a both student and site level variables.
Positive values for alpha (up to a max of role in student interest: F(3, 589) = 17.55, Goodness of fit (AIC and BIC) suggests that
1.00) indicate that there are greater differ- p <0.01. Pairwise comparisons in the same a parsimonious model with only signifi-
ences of opinion between learners. The table suggest that students completing cant predictors is a strong fit for these data.
observed values of 0.91 for the pre-inter- three AR field trips were significantly The results of the parsimonious model
vention and 0.93 for the post-intervention more interested in learning geoscience in (Table 3) indicate that there are three strong
GeoIS instrument indicate a high level of the future than students completing one or predictor variables for student interest
reliability (Murphy and Davidshofer, zero AR field trips.
1988). Given the established nature and
prior research conducted with the MSLQ, Table 3. Results from HLM modeling
we chose to use a confirmatory factor
analysis to assess instrument validity of Model 0 Model 1 Model 2 Model 3
the GeoIS. The fifteen GeoIS items complete parsimonious
coalesced onto a single factor based on 874 null model student level
observations with loadings ranging from
0.17 to 0.83. Based on this combination of Student Level
observations and loading values, the
adapted MSLQ instrument appears to Constant 15.10 0.97 5.78E-19 4.64E-18
measure a single construct at a significant
level (Stevens, 1999). The correlation GeoIS score pre-intervention 1.07 1.08 1.08
matrix (Data Repository Table S2 [see Gender 0.79 0.78
footnote 1]) revealed four statistically sig- Race 0.29 –0.55
nificant variables: (1) the pre-intervention
survey score; (2) institution; (3) STEM STEM major 101.94 3.58 3.09 2.18
major; and (4) number of AR field trips No. of ARFTs complete 2.00 1.47 1.72
completed. Despite a lack of statistical sig- Site classification 2.06 1.31
nificance, race and gender were kept as 65.98 65.09 65.47
Residual

Site Level

GeoIS score pre-intervention 1.50 1.71

AIC 5390.09 4169.36 4158.09 4169.52

BIC 5403.83 4208.78 4201.89 4200.20

AIC—goodness of fit; ARFTs—augmented reality field trips; BIC—goodness of fit; GeoIS—

geoscience interest survey; HLM—hierarchical linear modeling; STEM—Science, technology,

engineering, and mathematics.

www.geosociety.org/gsatoday 7