Emo · Methodology
Why Emo asks what it asks
The theories and measurement traditions behind each field of a check-in, what was deliberately left out, and what that means for how the data can be analysed.
Field names match the export (JSONL; the CSV writes
affect.valence as affect_valence).
0How firm each source is
Every source below carries one of three marks.
-
Evidence level
A
Established findings: they replicate, and there are meta-analyses.
-
Evidence level
B
An actively researched model, with criticism and competing theories.
-
Evidence level
C
A clinical tradition or a practical model without a strong experimental base. A source of vocabulary and procedure, never grounds for statistical conclusions.
1How data is collected
1.1Experience sampling and ecological momentary assessment
- Csikszentmihalyi & Larson (1987): the Experience Sampling Method (ESM). Shiffman, Stone & Hufford (2008): Ecological Momentary Assessment (EMA).
- The idea: many short measurements in the moment, in everyday surroundings, prompted by a random signal.
- Why: looking back (“how was your week?”) is distorted by peaks, endings and the present state (Kahneman’s peak-end rule; Robinson & Clore, 2002, the accessibility model: a retrospective report draws on beliefs about oneself, not on experience).
- For someone who has long been cut off from bodily signals, looking back is especially uninformative: beliefs about their own states are missing or formulaic.
What follows for the record
-
trigger = random: signal-contingent sampling. -
trigger = manual: event-contingent sampling. A growing share ofmanualentries means more spontaneous noticing, which is itself a target outcome. -
skipped: compliance. The EMA standard is at least 70–80% of prompts answered; below that the data is systematically biased, because the hard moments are the ones skipped (§5.3). - A short form (60 seconds or less): an EMA requirement for keeping compliance over weeks and months.
The response window is 15 minutes after the notification
appears. An answer is a tap on it, or a check-in started on
one’s own while it is still on screen. Later, the entry is
manual and the reminder counts as unanswered.
1.2Measurement reactivity
- Measuring changes what is measured (Barta, Tennen & Litt, 2012, a review of diary research).
- Here reactivity is the goal: turning attention to the body again and again is the intervention itself (affect labelling, §2.11; interoceptive training, §2.3).
- For analysis: a change in the state cannot be separated from a change in the skill of reading it. This is a fundamental limit; keep it in mind in every interpretation.
2The models behind the fields
2.1The circumplex model of affect
- Russell (1980); Russell & Barrett (1999), core affect.
- Any state can be placed on two continuous axes: valence (pleasant to unpleasant) and arousal (sluggish to activated). A named, discrete state is built on top of them.
- Why it matters: someone who cannot yet name what is going on can still rate pleasantness and energy. It is the lowest level of access, and the place to start.
-
For analysis: the (valence, arousal) coordinates are the
main raw variable. The word (
affect.word) is secondary.
2.2Constructed emotion and emotional granularity
- Barrett (2006, 2017, How Emotions Are Made). The competing position is basic emotions (Ekman, Panksepp). The debate is not settled; for keeping a diary it does not matter much.
- Granularity, or differentiation: telling close states apart (“anxiety” or “irritation” or “shame” rather than “bad”).
- Evidence level A: low granularity goes with poorer regulation, more impulsivity and alexithymia; high granularity with more adaptive strategies (Barrett et al., 2001; Kashdan, Barrett & McKnight, 2015, a review).
- The classic metric: the mean ICC between ratings of several states within a person across many moments; granularity = 1 − ICC (Tugade, Fredrickson & Barrett, 2004). It does not apply here directly: an entry holds one word, not ratings on a set of words. §5.1 lists stand-ins.
- The two-level list (8 broad words, then about 50 finer ones) is procedural scaffolding: someone with low granularity first picks a cluster, then refines it.
2.3Interoception
- Craig (2002, 2009): interoception as the basis of the subjective sense of self (the insula).
-
Garfinkel et al. (2015): a three-part model that matters
for analysis.
- Interoceptive accuracy: objective precision (heartbeat counting, for example).
- Interoceptive sensibility: one’s own rating of one’s sensitivity (the MAIA and BPQ questionnaires).
- Interoceptive awareness: the metacognitive match between the two.
affect.word_confidence. Accuracy cannot be measured by a diary; draw no conclusions about it. - Khalsa et al. (2018): a consensus review of interoception in psychiatry.
- Interoceptive training: Price & Hooven (2018), MABT, a body-awareness therapy based on mindfulness; randomised trials of moderate quality Evidence level B.
- A questionnaire for a periodic outside measurement (every 3–6 months): MAIA-2 (Mehling et al., 2018).
2.4Body maps
- Nummenmaa et al. (2014, PNAS, Bodily maps of emotions): topographies that hold across cultures. Fear and anxiety: the chest and upper body. Anger: the arms, head and chest. Sadness: less activation in the limbs. Depression: an overall “switching off”.
- For analysis: the spread of zones by valence or by word can be compared with these maps as a prior expectation. A mismatch is not an error; it is data.
2.5Somatic markers: the case for the whole idea
- Damasio (1994, Descartes’ Error); Bechara et al. (1997), the Iowa Gambling Task.
- Bodily signals take part in decisions before they reach awareness. Criticism: Dunn, Dalgleish & Lawrence (2006) find the evidence weaker than the popular version. The general idea (the body informs choice) holds; the details of the mechanism do not.
-
Relevance:
urgerecords a tendency to act, the closest available correlate of a “marker”.
2.6Action tendencies
- Frijda (1986): an affective state as readiness to act (action readiness).
- Lang (1995): motivational systems of approach and avoidance.
- The urge list (leave, shrink, freeze, move, hit, cry…) maps directly onto fight, flight and freeze, plus approach. It is useful precisely because an urge is often read earlier and more easily than a named state.
2.7Alexithymia: “nothing”, “don’t know”, “can’t tell”
- Sifneos (1973) coined the term. Bagby, Parker & Taylor (1994): TAS-20, with three factors: difficulty identifying feelings (DIF), difficulty describing feelings (DDF) and externally oriented thinking (EOT).
-
Thinking that is mostly operational and turned outward is
the pattern of living “outside oneself”: attention held on
tasks and the world rather than on inner states. EOT shows
up in
head_mode. - Valid “nothing” and “don’t know” answers are mandatory. Without them a person with alexithymia picks a random word, and the data looks better than it is.
- Outside measurement: TAS-20 at the start and every 6 months. Expected change: lower DIF and DDF.
- About half of autistic adults have alexithymia (Kinnaird, Stewart & Tsakanikos, 2019, a meta-analysis). Where this applies, the trajectory and the expectations differ (§6).
2.8Levels of emotional awareness: how the fields are ordered
- Lane & Schwartz (1987), the Levels of Emotional Awareness Scale (LEAS). The ladder: (1) bodily sensations, (2) action tendencies, (3) single named states, (4) blends of them, (5) blends in oneself and in someone else.
-
The record’s fields sit on exactly this ladder: the body
(1),
urge(2),affect.word(3). Several words or qualities in one entry mark a step up to level 4.
2.9Experiential avoidance: no “why”
- Hayes et al. (1996, 2006), ACT; the AAQ-II questionnaire (Bond et al., 2011).
- Avoiding inner experience through thinking, activity and analysis. Intellectualising is one form of it.
-
Mind-wandering: Killingsworth & Gilbert (2010,
Science), experience sampling with 2,250 people. About 47%
of the time, attention is not on the current activity, and
a wandering mind predicts lower happiness.
head_modeis a direct analogue of their question.
2.10Rumination and processing mode: no “why”, a short note
- Nolen-Hoeksema (1991, 2008): rumination as a factor in depression.
- Watkins (2008): constructive and unconstructive repetitive thought. The abstract mode (“why is this happening to me”, “what does it mean”) does harm; the concrete mode (“what exactly, where, how”) is neutral or helps.
- Kross & Ayduk (2011, 2017): self-distancing. Asking “why” from an immersed position intensifies affect without processing it.
-
Hence the note turns away causes and meanings (“because”,
“means”, “due to”, “that’s why”) and abstract-mode
questions (“why”, “what’s wrong with me”), and the form
asks “what and where”, never “why”. Hedges (“probably”, “I
think”, “it seems”) are not the abstract mode but caution,
close to
affect.word_confidence = guessand to the distanced stance of ACT.
Hedges are no longer turned away: “thinking about the deadline” is a fact about attention.
2.11Affect labelling: why the diary is itself an intervention
- Lieberman et al. (2007): naming affect lowers amygdala activity and raises ventrolateral prefrontal activity. Torre & Lieberman (2018), a review: the effect is robust and works as implicit regulation.
- Pennebaker (1997), expressive writing. Not used as a model: its protocol is long narratives with meanings and causes, which works against this purpose (§2.10). Only the finding is taken: putting affect into words regulates it.
2.12Titration and somatic approaches: 60 seconds, small doses
- Levine (1997, 2010), Somatic Experiencing: titration, pendulation, the window of tolerance (Siegel’s term, 1999). Ogden (2006), Sensorimotor Psychotherapy.
- Gendlin (1978), Focusing and the “felt sense”: the list of qualities (heavy, lump, tight…) comes from this tradition.
- The experimental base is weak; randomised trials of SE are few and small. It serves as a source of a safety procedure (do not flood) and of vocabulary, not as scientific grounds.
2.13Schema therapy: the “detached protector”
- Young, Klosko & Weishaar (2003). The Detached Protector mode: chronic disconnection from inner experience as protection, with the Emotional Deprivation schema as a typical root.
- Questionnaires: YSQ-S3, SMI. Not needed to analyse the diary; relevant for a therapist.
2.14Motivational design: no streaks, no achievements
- Deci & Ryan, Self-Determination Theory: external rewards undermine intrinsic motivation (Deci, Koestner & Ryan, 1999, a meta-analysis).
- For anyone who leans on a perfectionist, “in the head” mode, streaks turn the practice into one more task to complete, which reproduces the avoidance.
3What was deliberately left out
- PANAS and similar lists of 20 adjectives
- Long; demands distinctions that are not there yet; invites random picks.
- Emoji faces and mood trackers
- A label instead of the body; low granularity gets locked in.
- Polyvagal theory (Porges) as a foundation
- Popular in somatic circles, but its key neurophysiological claims are not supported (Grossman & Taylor, 2007; Grossman, 2023). The “ventral” and “dorsal” vocabulary is not used.
- HRV and wearable sensors
- They put exteroception in place of interoception. In Garfinkel’s terms sensibility is not accuracy, and sensibility is what needs training.
- Cognitive columns (the ABC form of CBT)
- They add a layer of thinking about thinking.
4Field, basis, metric
What each field of the export rests on, and what to compute from it.
trigger- BasisESM / EMA (§1.1)
-
Computethe share of
manualby week skipped- BasisEMA compliance (§1.1, §5.3), with the gaps
- Compute% answered; the pattern of gaps by time of day, day of week and context
latency_sec- Basisan indirect measure of what access costs, like reaction time in implicit tasks
- Computethe weekly median; its spread; latency by valence (longer towards the negative?)
body.locations- BasisNummenmaa (§2.4)
- Computezone frequencies; zones × word; zones × valence
body.qualities- Basisthe Focusing and SE vocabulary (§2.12); LEAS level 1 (§2.8)
- Computethe number of distinct qualities; the share of “nothing” and “can’t tell”
body.intensity- Basisnone of its own
- Computethe correlation with arousal. It is expected to be positive; its absence marks a split between body and affect
-
affect.valenceaffect.arousal - BasisRussell (§2.1)
- Computethe distribution; dynamic metrics (§5.2)
affect.word- Basisgranularity (§2.2); LEAS level 3
- Computedistinct words; entropy; the share of finer words against broad ones; the share of empty and “don’t know”
affect.word_confidence- Basisinteroceptive awareness (§2.3)
-
Computethe share of
sure; its change over time urge- BasisFrijda (§2.6); LEAS level 2
- Computethe distribution; urge × valence
head_mode- Basisavoidance and mind-wandering (§2.9)
- Computethe mean; its correlation with latency and with “nothing”
context_tag- BasisEMA context sampling
- Computeevery metric above, split by context
note- BasisWatkins, the concrete mode (§2.10)
- Computenothing: it is not analysed by numbers; it is material for therapy sessions
5Notes for analysis
5.1Stand-ins for granularity with one word per entry
The classic ICC method does not apply. Options:
- Distinct words in a window (a week, a month). Crude, but transparent.
- Shannon entropy of the word distribution in a window: H = −Σ pᵢ log pᵢ. Normalise by log(number of entries), or it grows with the amount of data.
- The share of finer words (the second level) among all chosen words.
- Conditional granularity: the number of different words within one (valence, arousal) cell. If (−2, 7) is always “anxiety”, it is low; if “annoyance”, “shame” and “wariness” turn up there too, it is growing.
- Optional, and not before 6 months: extend the form to 4–6 ratings for the words within a cluster, then compute the classic ICC.
5.2Dynamic affect metrics
- Inertia (emotional inertia): the lag-1 autocorrelation of valence (Kuppens, Allen & Sheeber, 2010). High inertia goes with depression and rigidity. As the channel recovers, inertia is expected to fall: the state starts responding to context.
- Instability: the MSSD or RMSSD of consecutive values (Jahng, Wood & Trull, 2008). Caution: entries are unevenly spaced, so use interval-adjusted variants, or only pairs of entries within one day.
- Variability: the within-person SD. A rise in SD early on is normal and expected (a signal appears); do not read it as getting worse.
- Separate within-day and between-day variance (a multilevel model: entries within days within weeks). For one person this is an ordinary two-level model, or simply aggregating by day.
5.3Gaps are not random
- In EMA, gaps fall more often on moments of strong affect or heavy load (missing not at random). Here a gap is also a likely marker of an avoidance episode.
- Analyse the pattern of gaps itself: time of day, day of week, the entry before (a gap after a negative entry?), context.
- Do not impute.
-
The cancellation reason in the reminder log
(
reason) separates technical gaps from a person’s:overdue(the server was down) andsettings(the window changed) are not gaps;tz_changeis a move;no_devicemeans notifications were off on every device (data about load too, but not a missed signal). Unanswered meanssentwith no entry carrying thatreminder_id. -
A planned pause in reminders (a day, three days, a week) is
an honest, analysable gap, not notifications quietly
switched off. Moments that fall into a pause stay in the
reminder log as
paused. They count neither as answers nor as gaps; analyse them separately, since the pause itself is data about load and avoidance. -
A partial entry (“Stop here”, or a form left with some
answers,
stopped_at) counts as a gap for compliance, and its answered fields are observed data, not imputation: they are never filled up to a full entry.
Cancelled reminders carry a reason; earlier
rows have none.
5.4Progress, and what only looks like it
The expected course by phase, qualitative, from clinical descriptions:
-
0–2 months
Many “nothing” and “don’t know” answers, high latency, high
head_mode. - 2–6 months Fewer “nothing” answers, more negative and more intense entries: not getting worse but access appearing. SD rises. Gaps may rise too, as avoidance pushes back.
- 6–12 months Positive valence appears; more distinct words, more manual check-ins, lower latency, lower inertia.
- Later Entries with several words or qualities at once (LEAS level 4).
Artefacts not to mistake for an effect
- Regression to the mean in the first weeks.
- Learning the interface: latency also falls simply as the form becomes familiar. Check: latency by type of answer (“nothing” against a specific word) should diverge if access is what changed, not familiarity.
- Season and day of week: valence on a Monday against a Saturday. Compare week with week, not day with day.
- Time-of-day bias: random signals should cover the whole waking window evenly; check that they do.
-
Changes to the instrument’s wording:
questions and pole labels are part of the measurement.
Every entry stores its wording version
(
ui.copy; no field means version 1). Compare slider values and words within one version only, and do not count a shift at a version boundary as an effect. Version 2 made the arousal poles neutral, which may weaken the link between valence and arousal.
Every hour of the window is drawn equally often when the window holds “number of reminders × the gap between them” (6 hours for three, 2 hours apart). Before that date the first and last hours were drawn about a third more often than the middle. A narrower window still works that way, and the skew is stronger there: in a 5-hour window the second and fourth hours are never drawn. Compare time-of-day shares within one period and one window.
5.5Minimum amounts
- 12–15 entries a week, at least, for weekly aggregates
- 60–100 entries for autocorrelation or MSSD, using only pairs less than 8 hours apart
- 8 weeks in each phase, at least, to compare before and after
- n ≥ 100 entries for correlations between fields (intensity × arousal, say); fewer make the confidence intervals useless
5.6Outside anchors, not from the diary
To check that the diary’s metrics reflect anything at all, every 3–6 months:
- TAS-20 alexithymia: DIF and DDF expected to fall ↓
- MAIA-2 interoceptive sensibility: Noticing, Emotional Awareness and Body Listening expected to rise ↑
- AAQ-II avoidance: expected to fall ↓
- PHQ-9, GAD-7 if needed, a check that the rise of negative entries in phase 2 has not turned into clinical depression or anxiety
All four are freely available and take 10–15 minutes together.
6Limits of the whole design
- N = 1. Each diary is an idiographic study: every conclusion is about change within the person who keeps it. Norms (how many words there “should” be) do not apply.
- Skill and state are mixed (§1.2): is a rise in recorded negative affect a rise in negative affect, or in the ability to notice it? The diary cannot tell. Only outside anchors and clinical context can.
-
If neurotype is the leading factor (autism,
or marked alexithymia as a trait), the goal shifts from
“restoring access” to “building an explicit system of
translation”: body, then words, then action. The metrics stay
the same; the expected pace and ceiling differ, and the share
of
affect.word_confidence = guessmay stay high, which is normal.
The diary does not replace a therapist. Somatic and affect-focused approaches (ISTDP, SE, EFT, schema therapy) rely on co-regulation in contact, which data cannot reproduce. The diary is a tool between sessions and a source of material for them.
7Key references
- Bagby, Parker, Taylor (1994). The twenty-item Toronto Alexithymia Scale. J Psychosom Res.
- Barrett (2006). Solving the emotion paradox. Pers Soc Psychol Rev.
- Barrett (2017). How Emotions Are Made.
- Barta, Tennen, Litt (2012). Measurement reactivity in diary research. In Handbook of Research Methods for Studying Daily Life.
- Bond et al. (2011). AAQ-II. Behav Ther.
- Craig (2002). How do you feel? Nat Rev Neurosci; (2009) Nat Rev Neurosci.
- Csikszentmihalyi, Larson (1987). Validity and reliability of the Experience-Sampling Method. J Nerv Ment Dis.
- Damasio (1994). Descartes’ Error.
- Deci, Koestner, Ryan (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychol Bull.
- Dunn, Dalgleish, Lawrence (2006). The somatic marker hypothesis: a critical evaluation. Neurosci Biobehav Rev.
- Frijda (1986). The Emotions.
- Garfinkel et al. (2015). Knowing your own heart. Biol Psychol.
- Gendlin (1978). Focusing.
- Grossman, Taylor (2007). Toward understanding respiratory sinus arrhythmia. Biol Psychol; Grossman (2023). Fundamental challenges and likely refutations of the five basic premises of the polyvagal theory. Biol Psychol.
- Hayes et al. (1996). Experiential avoidance and behavioral disorders. J Consult Clin Psychol; Hayes et al. (2006). Acceptance and Commitment Therapy: model, processes and outcomes. Behav Res Ther.
- Jahng, Wood, Trull (2008). Analysis of affective instability in ecological momentary assessment. Psychol Methods.
- Kashdan, Barrett, McKnight (2015). Unpacking emotion differentiation. Curr Dir Psychol Sci.
- Khalsa et al. (2018). Interoception and mental health: a roadmap. Biol Psychiatry CNNI.
- Killingsworth, Gilbert (2010). A wandering mind is an unhappy mind. Science.
- Kinnaird, Stewart, Tsakanikos (2019). Investigating alexithymia in autism: a systematic review and meta-analysis. Eur Psychiatry.
- Kross, Ayduk (2017). Self-distancing. Adv Exp Soc Psychol.
- Kuppens, Allen, Sheeber (2010). Emotional inertia and psychological maladjustment. Psychol Sci.
- Lane, Schwartz (1987). Levels of emotional awareness. Am J Psychiatry.
- Levine (2010). In an Unspoken Voice.
- Lieberman et al. (2007). Putting feelings into words. Psychol Sci; Torre, Lieberman (2018). Emotion Review.
- Mehling et al. (2018). MAIA-2. PLoS ONE.
- Nolen-Hoeksema, Wisco, Lyubomirsky (2008). Rethinking rumination. Perspect Psychol Sci.
- Nummenmaa et al. (2014). Bodily maps of emotions. PNAS.
- Ogden, Minton, Pain (2006). Trauma and the Body.
- Price, Hooven (2018). Interoceptive awareness skills for emotion regulation (MABT). Front Psychol.
- Robinson, Clore (2002). Belief and feeling: evidence for an accessibility model of emotional self-report. Psychol Bull.
- Russell (1980). A circumplex model of affect. J Pers Soc Psychol; Russell, Barrett (1999). J Pers Soc Psychol.
- Shiffman, Stone, Hufford (2008). Ecological Momentary Assessment. Annu Rev Clin Psychol.
- Sifneos (1973). The prevalence of “alexithymic” characteristics in psychosomatic patients. Psychother Psychosom.
- Tugade, Fredrickson, Barrett (2004). Psychological resilience and positive emotional granularity. J Pers.
- Watkins (2008). Constructive and unconstructive repetitive thought. Psychol Bull.
- Young, Klosko, Weishaar (2003). Schema Therapy.
Emo is a self-observation journal, not a medical device. This page is part of its source code (AGPL): github.com/podgorniy/emo
Contents ↑