ANALYSIS PLAN
Notes: We will conduct separate analyses for Sample 1 and Sample 2. We included a response check to
ensure that participants’ answers were effortful and honest. We will eliminate participants from the
analyses if they fail their response checks.
Power
Researchers may need larger sample sizes to detect small-to-moderate relationships among multiple
constructs in a factor model (Wolf, Harrington, Clark, & Miller, 2013). If the RISC, Romantic-RISC, and
Partner-RISC are related, we expect these associations would be weak-to-moderate. Wolf et al. (2013)
estimated that a three-factor model, with factors estimated to correlate at .30 and eight indicators per
factor, would be adequately powered ( = .81;  ≤ .05) with 160 participants. Given that an increased
number of indicators per factor typically reduces the required sample size (Wolf et al., 2013), we
reasoned that a minimum of 160 participants (per sample) would suffice for our three-factor model with
11 indicators per factor.
Based on previous interactions between RISC and self-esteem predicting interdependent behaviors,
which explained approximately 10-15% of the variance in interdependent behaviors overall (Baker &
McNulty, 2013), we conducted an apriori statistical power analysis using G*Power to determine a
sufficient sample size for examining comparable effects. Results suggested that a sample of at least 55
participants would be required to detect a small-to-medium effect (f2 = 0.15,  = .05,  = .20).
Basic Psychometric Properties
Factor Structure. We will examine scale properties separately for Sample 1 and Sample 2. The RomanticRISC and Partner-RISC scales are modifications of the original RISC scale, so they both contain the same
number of items as the original RISC (11 items total; reverse score items 8 and 9); however, they are
worded differently to tap their respective constructs. We will conduct a confirmatory factor analysis
(CFA) to test the model shown in Figure 1. This CFA model will examine two hypothesized properties of
the three scales: (1) we will test one-dimensional factor structures for each scale: RISC (replication),
Romantic-RISC, and Partner-RISC; and (2) we will test whether RISC, Romantic-RISC, and Partner-RISC
are unique constructs. First, for example, we aim to replicate the one-dimensional factor structure for
the RISC (Cross et al., 2000) by modeling associations between the 11 RISC items and one latent, RISC
construct. Second, to confirm that RISC, Romantic-RISC, and Partner-RISC are unique constructs, the 11
Romantic-RISC items and 11 Partner-RISC items should also load onto their respective, latent constructs.
Furthermore, the CFA covariance matrix should show covariance between the items that specify
different latent factors should be relatively closer to zero than the covariance between the items that
specify the same latent factor. That is, each RISC item should most strongly vary with other RISC items
compared to Romantic-RISC items and Partner-RISC items. Likewise, Romantic-RISC and Partner-RISC
items should most strongly vary with each other relative to the items for the other two latent variables.
Finally, although we expect the correlations to be weak, the CFA model in Figure 1 predicts that RISC,
Romantic-RISC, and Partner-RISC relate to each other. To test this hypothesis, we will compare the CFA
model in Figure 1 with an alternative CFA model that does not predict a relationship between the latent
constructs. The alternative model should better fit the data if the three constructs are independent. If
the three constructs are related, then the first model (Figure 1) should better fit the data. Support for
the first model would mean participants’ standing on one construct likely relates to their standing on the
other two constructs. For example, being defined by a romantic partner would increase the likelihood
that one is also defined by romantic relationships in general. By contrast, support for the second model,
would mean participants’ standing on one construct does not influence their standing on the other two
constructs. For example, being defined by a romantic-partner would not increase the likelihood that one
is defined by romantic relationships in general.
In the event that neither model fits the data well, we will consider modifications to covariances between
the latent constructs. For example, we would compare a third, alternative model if the data suggest two
of the latent constructs are related to each other (e.g., Romantic-RISC and Partner-RISC), but are not
related to the third construct (e.g., RISC). We will then replicate whichever model we retain in the
second sample.
Scale Reliability. We will examine coefficient alphas as estimates of each scales internal consistency.
Figure 1
*Note: number of observations = 561; parameters = 69; model df = 491
Convergent Validity
We will use CFA to examine the proposed correlations between measures, such as the hypothesized
associations between RISC and measures of interdependence and independence. For example, results
will support our hypothesis if the correlation between RISC and Interdependent Self-Construal is
significantly stronger than the correlation between RISC and Independent Self-Construal. To assess
whether the two correlations are significantly different, we would use the chi-square difference test,
which would allow us to compare two models; one model that constrains the correlations to be equal
compared to another model that does not constrain the correlations to be equal. The chi-square
difference test will support our hypotheses if the models are significantly different and the correlations
are in the hypothesized, positive direction.
Examination of Incremental Validity
In order to test our hypotheses that a construct uniquely predicts an outcome, controlling for another
construct, we will conduct regression analyses to examine the coefficient for said constructs. A construct
is uniquely predictive if it explains a significant amount of variance in the outcome, controlling for the
other hypothesized predictors in the model. For example, we predict that Romantic-RISC will explain
unique variance in the willingness to risk rejection to increase interdependence, controlling for
the tendency to be defined by close relationships (Hypothesis 11a). Thus, results would support
this hypothesis if Romantic-RISC significantly predicts the willingness to increase
interdependence when controlling for the significant influence of RISC as another predictor.
Replicating and Extending Previous RISC Interactions. To test interaction hypotheses—e.g., selfesteem will interact with RISC to predict various outcomes—we will model the influence of selfesteem, RISC, and the product of self-esteem X RISC on the outcome. Interaction occurs when
the interaction term significantly increases the amount of variance explained in the outcome.
To further test our hypotheses that the new constructs, such as Romantic-RISC, explain
romantic relationship outcomes better than RISC, we will compare the regression model that
includes the interaction between RISC and self-esteem to a model that includes the interaction
between Romantic-RISC and self-esteem. Results will support our hypotheses if modeling the
interaction between Romantic-RISC and self-esteem explains significantly more variance in
relationship outcomes compared to modeling the interaction between RISC and self-esteem.