Emulate: The conditional branching (only ask follow-up items if the participant says “yes” to a stressor). For perceived stress in a clinical/at-risk sample, this protects against burden on quieter moments and avoids forcing answers when nothing happened.
Reject: Forced selection of one stressor type. Perceived stress is about appraisal of overload and lack of control, not categorizing a single event. Two co-occurring small stressors can produce more felt stress than one big one, and the forced-choice item hides that.
The most relevant finding is that prompt frequency lowered participation interest (b = -0.56, p < .05) and prompt length raised effort (b = 1.01, p < .001). Since I am running 5x/day for 90 days — already heavy on the frequency dimension — I am holding the line on prompt length: 3–5 items, each ≤12 words, response time target under 30 seconds per ping. I am trading item count for compliance.
Borrow: From Example 1 (Maher et al., older adults’ PA), the single-screen, one-item-per-screen layout. For a clinical sample that may be cognitively taxed by stress itself, one item per screen reduces decision load.
Depart: From Example 3 (Adams et al., parental soothing), the nested flowchart with up to 7 follow-up items. That depth makes sense for a behavioral cascade, but for momentary stress appraisal it would balloon time and drop people. I cap at 5 items, no nested loops.
1. Visual Analog Slider (0–100).
2. Numeric Likert (0–10 NRS).
Best for state stress: The 0–100 slider. Continuous response matches the continuous nature of stress appraisal, gives more variance to model, and is sensitive to within-day change.
Perceived Stress
The degree to which individuals appraise situations in their lives as unpredictable, uncontrollable, and overloading their capacity to cope (Cohen, Kamarck, & Mermelstein, 1983).
Momentary Perceived Stress
The right-now appraisal of feeling overloaded, lacking control, or unable to cope. Treated as a state that can fluctuate within a day and across days, with both stable (trait) and time-varying (state) components.
Perceived Stress Scale, 10-item version (PSS-10). Cohen, S., Kamarck, T., & Mermelstein, R. (1983). A global measure of perceived stress. Journal of Health and Social Behavior, 24(4), 385–396.
Prompt: “The questions in this scale ask about your feelings and thoughts during the last month. In each case, please indicate how often you felt or thought a certain way.”
5-point frequency: 0 = never, 1 = almost never, 2 = sometimes, 3 = fairly often, 4 = very often. Items 4, 5, 7, 8 are reverse-scored. Total range 0–40.
Prompt: “Quick check-in. Please answer about right now.”
For momentary perceived stress, the highest-burden item is “Since the last check-in, did anything stressful happen?” paired with its follow-up “How much is it still affecting you right now?” This combination raises social-desirability pressure (people under-report stressors to seem fine) and adds time when the answer is yes. I kept it because reactivity — how a stressor is still being appraised — is the signal that distinguishes state from trait stress, and it lets me link momentary scores to events in the multilevel model. If pilot dropouts spiked, my fallback is to drop the yes-follow-up entirely and rely on the three slider items alone, treating event reactivity as a Phase 2 add-on once compliance is stable.
My EMA item “Right now, how stressed do you feel?” on a 0–100 slider maps to the PSS-10 item “In the last month, how often have you felt nervous and stressed?” on a 0–4 frequency scale. The slider captures within-person change in intensity at a moment, which the frequency scale cannot do — a monthly “sometimes” can hide both a flat 30/100 day and a bouncing 5-to-90 day. The slider also gives a continuous outcome that fits naturally into a multilevel model. What it still misses is the appraisal content — why this moment feels stressful — which is why item 4 (yes/no stressor) sits alongside.
For the item “Right now, how in control of what’s happening do you feel?”, the participant has to (1) appraise their current situation, (2) retrieve recent events from the last hour or so, (3) compute a sense of control aggregating those, and (4) translate that into a slider position. I’m assuming people can introspect on perceived control with reasonable accuracy in a few seconds. If that assumption is wrong — say, in a clinical sample where dissociation or rumination distorts introspection — the item may collapse into general mood rather than control specifically. A fallback would be to add a concreteness anchor (“Compared to your usual…”) to reduce drift.
For a clinical/at-risk participant, I expect 9am responses to start moderate as the day’s demands come into view, climb through 12pm and 3pm (mid-day cognitive load and accumulated stressors), peak around 3–6pm, and decline by 9pm. The slider items should show the most within-person variability mid-day; the yes/no stressor item is likely to floor in the morning (nothing has happened yet) and ceiling in the late afternoon on bad days. The control item is the one I’d worry about hitting a low-end floor in a high-symptom subgroup, which would limit variance and pull down my random slope estimates.
I’d fit a multilevel model with momentary stress (slider 1) as the outcome and time-of-day plus item-4 (recent stressor) as level-1 predictors, with random intercepts and random slopes by person. The key decision is centering: I’d use person-mean centering for the recent-stressor predictor rather than grand-mean centering. Person-mean centering separates within-person reactivity (does this moment with a stressor feel worse than my average?) from between-person trait differences (do people with more frequent stressors carry higher overall stress?), and the within-person estimate is what the EMA design was built to recover. Grand-mean centering would conflate the two and waste the time-varying signal I designed the instrument to capture.