3. Apropiación mediática
3.4. Factores que Influyen en la apropiación mediática
3.4.1. Accesibilidad, actitud, competencia y uso didáctico de las TIC en la docencia
To test hypotheses (i) and (iii), the first GLM (hereby called GLM-1) predicted the binomial outcome correct/incorrect (0/1), with a logit link and the predictor and covariate variables1listed in Table 5.2:
Table 5.2: GLM-1 Predictor Variables
Predictor/Covariate Code Levels Reference
(SR)Training-Intuition sr.tr.intuit true=1/false=0 false
(SR)Training-Rule sr.tr.rule true=1/false=0 false
(SR)Training-Memorized sr.tr.mem true=1/false=0 false
(SR)Testing-Intuition sr.te.intuit true=1/false=0 false
(SR)Testing-Rule sr.te.rule true=1/false=0 false
(SR)Testing-Memorized sr.te.mem true=1/false=0 false
Target Pattern target.pattern c=1/ncs=2/ncns=3 c
Learning Condition learning.condition E=0/I=1 E
*Covariates areitalicized
GLM-1 Equation
Correct∼sr.tr.intuit∗target.pattern+sr.tr.rule∗target.pattern
+sr.tr.mem∗target.pattern+sr.te.intuit∗target.pattern
+sr.te.rule∗target.pattern+sr.te.mem∗target.pattern
+target.pattern∗learning.condition
In addition, each of the two-way interactions among self-reported strategies∗target pattern were included in GLM-1 (e.g.sr.te.intuit:target.pattern-c), as was an interaction term for learning condition
and target pattern. The formula used forgeeglmis given above. Model parameters were estimated using GEE with clusters corresponding to participant ID as a factor (118 total levels) to account for within-participant correlations for repeated measures.
The purpose of this model was to test for effects of self-reported learning strategies on partici- pants’ performance conditioning with respect to the target pattern type and learning conditions. The estimated coefficients are given in Table 5.3 below:
1
Here, I use “covariate” to refer to a categorical variable that, unlike the task and pattern conditions, was not manipulated, but was rather measured via post-experiment questionnaires.
Table 5.3: GEE Coefficient Estimates for GLM-1
GEE Results
Coefficient Estimate Std.err Wald Pr(>|W|)
(Intercept) -0.3283 0.1060 9.6000 0.0019 ** (SR)-Training-Intuition 0.1375 0.0932 2.1800 0.1400 NCNS 0.0449 0.1751 0.0700 0.7978 NCS 0.3760 0.1830 4.2200 0.0399 * (SR)-Training-Rule 0.2422 0.0997 5.9000 0.0151 * (SR)-Training-Memorized 0.2536 0.0983 6.6600 0.0099 ** (SR)-Testing-Intuition 0.0023 0.0902 0.0000 0.9797 (SR)-Testing-Rule 0.0379 0.1055 0.1300 0.7192 (SR)-Testing-Memorized -0.0828 0.0842 0.9700 0.3255 Cond-I 0.1064 0.0849 1.5700 0.2100 (SR)-Training-Intuition:NCNS 0.3455 0.1622 4.5400 0.0332 * (SR)-Training-Intuition:NCS 0.0083 0.2106 0.0000 0.9686 (SR)-Training-Rule:NCNS 0.1480 0.1385 1.1400 0.2852 (SR)-Training-Rule:NCS -0.3328 0.1906 3.0500 0.0808 . (SR)-Training-Memorized:NCNS -0.4229 0.1449 8.5200 0.0035 ** (SR)-Training-Memorized:NCS -0.3381 0.1748 3.7400 0.0531 . (SR)-Testing-Intuition:NCNS -0.2781 0.1998 1.9400 0.1641 (SR)-Testing-Intuition:NCS -0.0726 0.1866 0.1500 0.6973 (SR)-Testing-Rule:NCNS -0.3350 0.1702 3.8700 0.0491 * (SR)-Testing-Rule:NCS -0.0471 0.1859 0.0600 0.8002 (SR)-Testing-Memorized:NCNS 0.2772 0.1663 2.7800 0.0956 . (SR)-Testing-Memorized:NCS 0.0889 0.1355 0.4300 0.5120 NCNS:Cond-I -0.3553 0.1204 8.7200 0.0032 ** NCS:Cond-I -0.2803 0.1476 3.6100 0.0575 . Signif. codes: [*** = 0.001, ** = 0.01, * = 0.05, . = 0.1]
Looking at the non-interaction coefficients for each learning strategy, we see a significant positive effect for(SR) Training-Rule(p = 0.0151*), indicating that participants who reported seeking a rule in training were more likely to answer correctly in the testing portion of the experiment than those who didn’t. Similarly, a positive effect is found for(SR) Training-Memorized(p= 0.0099**) suggesting that trying to “memorize” the training words interestingly lent an advantage in testing even though the design of the experiment should have made simple memorization useless since each word would only appear once. Although, if participants were somehow “accessing” memorized forms and comparing novel forms to such forms, this might account for the positive effect. On the other hand, no significant effect is found for(SR) Training-Intuition(p = 0.1400), suggesting no overall advantage to be gained from implicit learning, although we do still see a non-significant positive effect.
From the self-reported learning-strategy data, it seems that participants’ strategies in the testing phase had very little effect on their performance (although we see a significant interactions of the NCNSpattern with rule-users and memorizers). On the other hand, we see several significant effects linked totrainingphase strategies. This might indicate that the process by which alternations are learnedis more linked to participant performance than the means by which the acquired knowledge is accessed after the fact.
Considering the non-interactions coefficients for target patterns, we see a significant positive effect of being assigned to aNCSpattern condition on the likelihood of being correct (p= 0.0399*), while no significant effect is observed for theNCNStarget pattern.
With GLM-1, we are particularly interested in the contrast between performance for explicit and non-explicit learners for target patternC. This contrast was tested by generating least-squares means for the interaction terms(SR) Training-Rule=true:C=trueand(SR) Training-Rule=false:C=true, and testing the resulting odds ratio for significant positive divergence from 1.
Table 5.4: GLM-1, Comparison of Explicit and Non-Explicit Learners for PatternC Least-Squares Means
term lsmean SE(α=0.05) LCL UCL
(SR) Training-Rule=false:C -0.1008 0.05 -0.20497 0.00328 (SR) Training-Rule=true:C 0.1413 0.07 -0.00361 0.28623
contrast odds.ratio SE(α=0.05) z.ratio p.value
(SR) Training-Rule=true:C – (SR) Training-Rule=false:C 1.27389 0.08 -2.429 0.0151
As shown in Table 5.4, we find an odds ratio of1.274for the contrast of the two interactions, which shows that the likelihood of being correct is significantlygreaterfor explicit learners in theC
pattern conditions vs. non-explicit learners (p= 0.0151*). This seems to fall in line with hypothesis (i) which predicted a strong advantage for explicit learners when dealing with opaque patterns like theCpattern. However, we also know that adults generally favor explicit learning, and so it would be helpful to contrast explicit learners and non-explicit learners for the other pattern conditions to insure that this isn’t just an across-the-board advantage for explicit learners. The results of this contrast analysis is given in Table 5.5 below:
Table 5.5: GLM-1, Comparison of Explicit and Non-Explicit Learners for all Patterns
Least-Squares Means
term lsmean SE(α=0.05) asymp.LCL asymp.UCL
(SR) Training-Rule=false:C -0.10 0.05 -0.20 0.00 (SR) Training-Rule=true:C 0.14 0.07 -0.00 0.29 (SR) Training-Rule=false:NCNS -0.44 0.12 -0.68 -0.20 (SR) Training-Rule=true:NCNS -0.05 0.07 -0.18 0.08 (SR) Training-Rule=false:NCS -0.05 0.11 -0.25 0.16 (SR) Training-Rule=true:NCS -0.14 0.09 -0.31 0.04
contrast odds.ratio SE(α=0.05) z.ratio p.value
(SR) Training-Rule=false:C – (SR) Training-Rule=true:C 0.78 0.08 -2.43 0.02 (SR) Training-Rule=false:NCNS – (SR) Training-Rule=true:NCNS 0.68 0.07 -4.06 0.00 (SR) Training-Rule=false:NCS – (SR) Training-Rule=true:NCS 1.09 0.18 0.56 0.58 (SR) Training-Rule=true:C – (SR) Training-Rule=true:NCNS 1.21 0.12 1.91 0.06 (SR) Training-Rule=true:C – (SR) Training-Rule=true:NCS 1.32 0.15 2.40 0.02 (SR) Training-Rule=true:NCNS – (SR) Training-Rule=true:NCS 1.09 0.12 0.77 0.44
In Table 5.5, we see a more complete set of contrast tests between explicit and non-explicit learners across the various target pattern conditions. As before, we see the significant advantage of explicit-learners over non-explicit learners in theCpattern conditions (p= 0.02*). We also see an even greater(!) difference between explicit learners and non-explicit learners in theNCNSpattern conditions (p <0.0001), raising doubt that the explicit advantage in theCcondition was a product of the target pattern. However, we see no significant difference in the performance of explicit vs.
non-explicit learners in theNCSpattern conditions, showing that the explicit advantage is at least not universal.
Contrasting the performance of explicit learners in theCpattern condition vs. in the non-circular conditions shows, we find a marginally significantly greater likelihood of being correct for(SR) Training-Rule=true:Cparticipants over both(SR) Training-Rule=true:NCNS(p= 0.06) and(SR) Training-Rule=true:NCS(p= 0.02*) participants. This is admittedly somewhat surprising given that theNCSandNCNSpatterns are structurallysimplerthan theCpattern.
It is difficult to draw a simple conclusion from the results of GLM-1 with regard to hypothesis (i). Although we do find a significant advantage for self-reported rule-seekers inCpattern conditions over non-rule-seekers, this advantage also arises in theNCNSpattern conditions. Comparing the effects of rule-seeking across pattern categories does show, however, that it has the greatest effect on learning outcomes in theCpattern conditions.
GLM-1 also can shed light on hypothesis (iii) which is that substantive bias in a dual-system model like that proposed in this thesis (§3.1) may be an exclusive property of the implicit learning mechanism. That is, if theNCSandNCNStarget patterns are substantively motivated and substan- tively non-motivated respectively, then hypothesis (iii) predicts a significantly greater positive effect of implicit learning in theNCSconditions than in theNCNS.
Referring back to the coefficient estimates table for GLM-1, Table 5.3, we see a significant positive effect for(SR)-Training-Intuition:NCNS(p=.0332) and no significant effect of the(SR)- Training-Intuition:NCSinteraction. This seems to stand in direct contrast to hypothesis (iii), although the existence of a strong distinction between the two interactions keeps us from needing to rule out substantive bias, as it may be the case that what has been deemed the substantively motivated pattern may in fact be less substantively motivated than the proposed non-motivated pattern.
In summary, GLM-1 evaluates the effects of self-reported learning strategies, learning condition, and their interactions with target patterns on the likelihood of making a correct answer. We seem to find support for hypothesis (i) from this data given that there is a significant advantage for rule-seekers in theCpattern conditions over non-rule seekers acquiring theCpattern,andover rule-seekers in theNCNSandNCSconditions, suggesting that theC pattern may be more conducive to explicit learning. On the other hand, GLM-1 does not seem to support hypothesis (iii), for we fail to find an
advantage for theNCSpattern over theNCNSin interaction with implicit learning. Instead, GLM-1 suggests that implicit learners faired better with theNCNSpattern than theNCSpattern.