SEM Speed-run

1. Develop hypotheses to represent your model

Direct Effects
H1a. Usefulness has a positive effect on Information Acquisition.
Logic goes here
H1b. Usefulness has a positive effect on Decision Quality.
Logic goes here
Mediated Effects
H2a. Usefulness mediates the relationship between Anxiety and Decision Quality.
H2b. Usefulness mediates the relationship between Playfulness and Decision Quality.

Moderated Effects
H3a. Experience dampens the negative relationship between Anxiety and Decision Quality.
H3b. Experience strengthens the positive relationship between Playfulness and Information Acquisition.

Multigroup Effects
H4a. The positive relationship between Playfulness and Usefulness is stronger for females.
H4b. The negative relationship between Anxiety and Usefulness is stronger for females.

2. Case Screening
1. Missing data in rows

We removed two rows in our dataset due to missing data over 20%.

2. Unengaged responses

We removed two cases due to being not engaged (they answered somewhat agree to every likert
scale item).

3. Outliers (on continuous variables)

There was one outlier in terms of age. The repsondentn reported being 9 years old. As these
are college students, we replaced the value with 19, assuming typographical input error.

3. Variable Screening
1. Missing data in columns

We observed two missing values in playful4. We looked at the surrounding values of the other
indicators for the latent factor Playfulness, and we used mode value for that respondent to impute
the missing values.

2. Skewness & Kurtosis

We observed fairly normal distributions for our indicators of latent factors, and for all other variables
(e.g., age, experience) in terms of skewness. However, we observed mild kurtosis for the indicators
of our dependent varialbes (IQ and DQ) and for one of our mediators (Usefulness). These kurtosis
values ranged from benign to 2.5. While this does violate strict rules of normality, it is within more
relaxed rules suggested by Sposito et al (1983) who recommend 3.3 as the upper threshold for
normality.

4. Exploratory Factor Analysis

Removed: InfoAcq5, DecQual11, DecQual13 (low loadings and as indicated by reliability analysis)

Pattern Matrixa
Cronbach’s Factor
Alpha 0.942 0.934 0.914 0.908 0.819 0.841
InfoAcq_1 .740
InfoAcq_2 .724
InfoAcq_3 .703
InfoAcq_4 .570
DecQual_6 .609
DecQual_7 .905
DecQual_8 .999
DecQual_9 .746
DecQual_10 .736
DecQual_12 .558
Useful_1 .782
Useful_2 .816
Useful_3 .946
Useful_4 .955
Useful_5 .848
Useful_6 .726
Useful_7 .677
Anxiety_1 .640
Anxiety_2 .864
Anxiety_3 .921
Anxiety_4 .903
Anxiety_5 .911
Anxiety_6 .819
Anxiety_7 .646
CompUse_1 .740
CompUse_2 .766
CompUse_3 .695
CompUse_4 .682
Playful_1 .601
Playful_2 .837
Playful_3 .777
Playful_4 .683
Playful_5 .903
Playful_6 .933
Playful_7 .642
Extraction Method: Maximum Likelihood.
Rotation Method: Promax with Kaiser Normalization.
a. Rotation converged in 6 iterations.

Factor Correlation Matrix

Factor 1 2 3 4 5 6
1 1.000 -.148 .342 .538 .270 .523
2 -.148 1.000 -.287 -.080 -.177 -.145
3 .342 -.287 1.000 .273 .389 .429
4 .538 -.080 .273 1.000 .163 .622
5 .270 -.177 .389 .163 1.000 .297
6 .523 -.145 .429 .622 .297 1.000
Extraction Method: Maximum Likelihood.
Rotation Method: Promax with Kaiser Normalization.

1. Iterate until you arrive at a clean pattern matrix

2. Adequacy (KMO, Communalities)
3. Convergent validity (loading amplitude on pattern matrix)
4. Discriminant validity (No major cross loadings or correlations)
5. Reliability (cronbach’s alpha)
5. Confirmatory Factor Analysis

Removed the following items due to model fit discrepancies (inflating chi-square): Playful5,
Usefulness4. We felt justified in doing this as these items belonged to large latent reflective factors
and were thus somewhat redundant.
 1. Obtain a roughly decent model quickly (cursory model fit, validity)
MS CompUs
CR AVE MaxR(H) Useful Anxiety Playful DecQual InfoAcq
V e

Useful 0.929 0.687 0.323 0.938 0.829

Anxiety 0.935 0.677 0.078 0.953 -0.136 0.823

Playful 0.897 0.594 0.192 0.908 0.345 -0.279 0.771

DecQual 0.910 0.630 0.478 0.923 0.566 -0.081 0.259 0.794

CompUs
0.819 0.531 0.167 0.821 0.310 -0.183 0.408 0.166 0.729
e

InfoAcq 0.843 0.574 0.478 0.844 0.569 -0.150 0.438 0.691 0.300 0.758

2. Do configural, metric, and scalar invariance tests (if using grouping variable in causal
model)

Configural invariance good (as evidenced by good model fit measures when estimating two
groups freely - i.e., without constraints)

Metric invariance was good as evidenced by a non-significant chi-square difference test

between the unconstrained and fully constrained models where the regression weights were
constrained.

Scalar invariance was only partially met due to the need to unconstrain all but two of the
items’ intercepts for Anxiety and two items for playfulness and one item for compuse.

3. Validity and Reliability check

We observed convergent and discriminant validity as evidenced by (convergent is AVE

above .5, discriminant is square root of AVE greater than coorelations) and reliability
(evide3nced by the CR value above 0.700)

4. Common method bias (marker if possible, CLF either way)

We identified significant method bias in our model as indicated by a significant chi-square difference
test between a zero-constrained and unconstrained model. So, we will control for the SocDes factor
in our structural analyses.

5. Final measurement model fit

 Measure Observed Threshold
Chi-square
DF
CFI .931
RMSEA .045
PCLOSE .190
SRMR 0.040

6. Optionally, impute factor scores

6. Structural Models
1. Multivariate Assumptions
1. Outliers and Influentials

We ran a cook’s distance analysis to determine if any (multivariate) influential outliers existed. In no
case did we observe a cook’s distance greater than 1. Most cases were far less than 0.100.

2. Multicollinearity

We examined variable inflation factors for all predictors on our dependent variables and observed no
VIFs greater than 2, which is far less than the threshold of 10.

2. Include control variables in all of the following analyses

Direct Effects
H1a. Usefulness has a positive effect on Information Acquisition.
Evidence: Beta = 0.465*** (p<0.001)
H1b. Usefulness has a positive effect on Decision Quality.
Evidence: Beta = 0.549*** (p<0.001)

Mediated Effects
H2a. Usefulness mediates the relationship between Anxiety and Decision Quality.

Not supported

User-defined estimands: (Group number 1 - Default model)

 Estimat
Parameter Lower Upper P
e
AxB -.0215 -.0918 .0436 .4981

H2b. Usefulness mediates the relationship between Playfulness and Decision Quality.
YES!

User-defined estimands: (Group number 1 - Default model)

Estimat
Parameter Lower Upper P
e
AxB .1776 .1096 .2738 .0006

Moderated Effects
H3a. Experience dampens the negative relationship between Anxiety and Decision Quality.
Not supported (nonsignificant beta from IV and from interaction)
H3b. Experience strengthens the positive relationship between Playfulness and Information Acquisition.
Not supported (nonsignificant beta from IV and from interaction)

Multigroup Effects
H4a. The positive relationship between Playfulness and Usefulness is stronger for females.
H4b. The negative relationship between Anxiety and Usefulness is stronger for females.
We ran a chi-square difference test with the unconstrained vs constrained (only individual path) models
and found no significant difference. Not supported. Beta =.29male .34 female