Quantifying Emotional Flow: Testing the Emotional Flow Hypothesis from a Longitudinal Latent Growth Curve (LGC) Modeling Approach

ABSTRACT

This paper presents a longitudinal, latent growth curve (LGC) modeling approach to refine the emotional flow measure and hypothesis testing. Emotional flow is operationalized as the marked within-individuals variations in one or more discrete emotions over time, which can be modeled as the amount and shape of change in emotions during message exposure. Emotional flow effects are tested in the LGC framework using data collected from a web-based experimental study where individuals (US Qualtrics Panel, N = 620) read an anti-sugary sweetened beverage message in the standard threat appeal format. Simultaneous fear and hope flows were established with unconditional LGC modeling. The two flows and their interaction were then used to predict message effects outcomes. Results showed that flow effects were nonsignificant when either the fear flow or the hope flow was relatively flat in form, but robust when both emotional flows were with marked variations over the course of message exposure.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

1. This does not mean multiple emotional flows can’t be in the same valence, for example, a fear flow along with a disgust flow can co-occur.

2. The stimuli messages and the data files are available at: https://osf.io/7qn3w/

3. There was a Heywood case in the initial model: The error term of Fear@Time1 was −0.263, with a standard error of 1.349, which means (1) the error term was not significantly different from 0, and (2) the negative value of the error variance was probably due to sampling error (Kolenikov & Bollen, 2012). To address that issue, the error term for Fear@Time1 was fixed at a small positive value (0.0005).

4. The first observed indicator as 5.71*fear@Time1*1.56*hope@Time1 and the other as 1.0*fear@Time2*1.0*hope@Time2. The other two observed emotion measures, fear@Time0 and hope@Time0, dropped out since the Shape factors had loadings of 0 on them (see, Bollen & Paxton, 1998 for the specification of latent interaction terms).