What it is
Subjective wellbeing (SWB) and non-cognitive skills are distinct but related constructs that capture dimensions of human welfare and capability not measured by income, consumption, or test scores.
Subjective wellbeing encompasses three components: life satisfaction (a cognitive evaluation of how one’s life is going overall), positive affect (the frequency of positive emotions in daily life), and negative affect (the frequency of negative emotions). Measurement draws on validated scales — the Cantril Ladder for life satisfaction, the Positive and Negative Affect Schedule (PANAS), and the Patient Health Questionnaire (PHQ) for depression symptoms — that have been widely used across cultural contexts.
Non-cognitive skills are psychological traits that shape how people behave in academic, labour, and social settings: self-efficacy (belief in one’s ability to achieve goals), locus of control (belief in whether outcomes are determined by one’s own actions or external forces), grit (perseverance and passion for long-term goals), and growth mindset (belief that abilities can be developed). Heckman, Stixrud and Urzua (2006) establish that non-cognitive skills predict earnings, health, and criminal behaviour as strongly as cognitive ability in US data; subsequent work has extended these findings to developing country contexts.
When to use it
SWB and non-cognitive skills are appropriate outcomes when: the study evaluates an intervention that directly targets psychological dimensions of wellbeing or capability (mental health programmes, social protection, economic empowerment); when these measures are plausible mediators of behavioural treatment effects; or when the research aims to characterise what a programme does for participants beyond economic outcomes.
Haushofer and Shapiro (2016) measure subjective wellbeing and psychological outcomes — depression, stress, cortisol — as primary outcomes in a cash transfer experiment in Kenya, finding substantial effects on both economic and psychological dimensions of welfare. Blattman, Jamison and Sheridan (2017) study a cognitive behavioural therapy programme in Liberia and use self-efficacy and locus of control as primary outcomes. Banerjee et al. (2015) include happiness and life satisfaction measures across six countries in a graduation programme evaluation and find effects in some sites but not others, illustrating both the sensitivity and the cross-site comparability of these measures.
How it works
Life satisfaction. The Cantril Ladder asks respondents to imagine a ladder with the best possible life at the top (10) and the worst possible life at the bottom (0), and to indicate which rung they currently stand on. Single-item, takes under a minute, internationally comparable. The World Happiness Report uses it as its primary measure.
Depression and mental health. The PHQ-9 (nine items, each scored 0–3) screens for depression symptoms over the past two weeks. The PHQ-2 is a two-item shortform used as a screener. Both are validated for use in low-income country settings. The Center for Epidemiologic Studies Depression Scale (CES-D) is a widely used 20-item alternative. Scores above a threshold are used to classify probable depression; the continuous score is used in regression analysis.
Positive and negative affect. The PANAS scale asks about 20 emotions experienced in a reference period (today, this past week). Responses generate a positive affect score and a negative affect score. Shorter versions (PA-NAS-10) are available for field use.
Self-efficacy. The General Self-Efficacy Scale (Schwarzer & Jerusalem, 1995) is a 10-item scale asking respondents to rate their agreement (1–4) with statements like “I can always manage to solve difficult problems if I try hard enough.” The scale has been validated in over 25 countries. Shorter domain-specific adaptations — self-efficacy for farming, for health behaviour, for employment — are also used.
Locus of control. The Rotter (1966) scale or shorter adaptations ask respondents to choose between paired statements — “What happens to me is my own doing” vs. “Sometimes I feel I don’t have enough control over the direction my life is taking.” External locus of control (believing outcomes are determined by luck or powerful others) is associated with lower economic outcomes and is potentially modifiable by interventions that increase agency.
Grit. The Grit Scale (Duckworth et al., 2007) is a 12-item (or 8-item short form) scale measuring consistency of interests and perseverance of effort. Validated primarily in US student populations; cross-cultural validation in developing country contexts is less established.
Key decisions
Selecting scales for the context. Not all validated scales perform equally in all contexts. Items requiring abstract self-reflection (“I feel I have a number of good qualities”) may produce different response patterns in populations with limited education or different norms around self-presentation. Pilot testing each scale with a small convenience sample, checking inter-item correlations and Cronbach’s alpha, and running cognitive interviews to identify confusing items are important steps before committing to a scale in the main survey.
Translation and back-translation. Psychological scales require careful translation — not just linguistic translation but conceptual adaptation. The instruction to back-translate and check that the back-translation matches the original meaning is standard (see the Survey Translation guide). Items referring to emotional states or personal traits are particularly vulnerable to translation errors that produce non-equivalent items across language versions.
Reference period. Life satisfaction questions ask about now, while affect questions typically specify a reference period (today, the past week, the past month). The choice of reference period affects both the mean and variance of responses: “today” is noisier than “the past month” but more sensitive to recent events. The reference period should match the theoretical construct — if the intervention is expected to affect daily mood, a shorter reference period is more appropriate.
Acquiescence bias. Respondents in many cultural contexts show a tendency to agree with positively worded statements regardless of content (“yes-saying”). This inflates scores on scales with predominantly positive items. Reverse-coded items — where agreement indicates a negative response — help identify and correct for acquiescence bias. Balanced scales with roughly equal numbers of positively and negatively coded items are preferred.
Order effects. Asking about life satisfaction after a negative affect battery, or after a survey module about illness or hardship, systematically lowers satisfaction scores. SWB measures should be placed early in the questionnaire, before question blocks that might prime specific emotional states.
Caveats & common mistakes
Conflating components. Life satisfaction, positive affect, and negative affect are empirically correlated but conceptually distinct and should not be summed into a single SWB score without justification. An intervention can increase positive affect without changing life satisfaction. Reporting each component separately and noting their correlations is more informative than collapsing them.
Reference point effects. Self-reported wellbeing is sensitive to what is salient in the respondent’s environment at the moment of measurement. Seasonal effects (surveys conducted during harvest vs. lean season), recent shocks (illness, rainfall failure), and questionnaire context all shift responses. Panel designs that track the same respondent over time can difference out fixed determinants, but transient contextual effects remain a concern.
Measurement equivalence across groups. Comparing SWB or non-cognitive skill scores across groups — treatment and control, or male and female — assumes that the same scale measures the same construct with the same metric in both groups. If item loadings or factor structures differ across groups, raw score comparisons are invalid. Testing for measurement invariance is important when cross-group comparisons are a primary analysis.
Non-cognitive skills and the Heckman critique. Non-cognitive skills are measured by self-report and may themselves be affected by the conditions being studied. A programme that increases income might also increase self-efficacy because respondents feel better about themselves when wealthier, without actually changing their underlying capacity or behaviour. Instruments that affect skills but not income (or vice versa) are needed to establish that skills independently affect outcomes, not just that they are correlated with them.
Analysis Guide
# PHQ-9 depression score (9 items, each 0-3): phq_1 to phq_9
import pandas as pd
import numpy as np
import statsmodels.formula.api as smf
import pingouin as pg
# 1. Sum PHQ-9 items into a total score and flag probable depression — the standard
# clinical cut-point of 10 captures moderate-to-severe symptoms over the past two weeks
phq_cols = [f"phq_{i}" for i in range(1, 10)]
df["phq9_score"] = df[phq_cols].sum(axis=1)
df["probable_depression"] = (df["phq9_score"] >= 10).astype(int)
# 2. Inspect the Cantril ladder distribution — single-item 0-10 life satisfaction;
# population means in low-income contexts typically fall between 4 and 6
print(df["cantril_ladder"].describe())
# 3. Reverse-code items 2 and 4 of the self-efficacy scale, then compute the mean —
# reverse coding aligns all items so higher values consistently indicate higher
# self-efficacy; using the mean keeps the result on the original 1-4 scale
df["se_item2"] = 5 - df["se_item2"]
df["se_item4"] = 5 - df["se_item4"]
se_cols = [f"se_item{i}" for i in range(1, 11)]
df["self_efficacy"] = df[se_cols].mean(axis=1)
# 4. Internal consistency — Cronbach's alpha should exceed 0.70 for a usable scale;
# below 0.60 means items are not measuring the same construct reliably, which
# typically signals translation problems or unsuitable items for this context
alpha, _ = pg.cronbach_alpha(data=df[se_cols])
print("Cronbach alpha:", alpha)
# 5. Regress life satisfaction and depression score on treatment — positive treatment
# coefficient on cantril_ladder = higher life satisfaction; negative coefficient on
# phq9_score = fewer depression symptoms (an improvement)
fit_cantril = smf.ols("cantril_ladder ~ C(treatment) + age + C(female) + log_hh_expenditure",
data=df).fit(cov_type="HC1")
fit_phq = smf.ols("phq9_score ~ C(treatment) + age + C(female) + log_hh_expenditure",
data=df).fit(cov_type="HC1")
print(fit_cantril.summary()); print(fit_phq.summary())XLSForm / SurveyCTO
Display each scale item as a select_one question using a Likert-format choice list. Use a separate choice list for each scale to avoid confusion from different response options. Reverse-coded items should be identified in the codebook but NOT reverse-coded in the instrument — recode in analysis to prevent enumerator confusion. Pre-test all items with cognitive interviews in the study language. For the PHQ-9, display the standard two-week reference period in the module header rather than repeating it in each item. Include the standard SurveyCTO validation to prevent missing values on required items.
Reading the output
phq9_scoreranges from 0 to 27. The standard clinical thresholds are: 0–4 minimal, 5–9 mild, 10–14 moderate, 15–19 moderately severe, 20–27 severe depression.probable_depression(score ≥ 10) flags moderate-to-severe cases and is the standard binary outcome used in development economics programme evaluations.cantril_ladderranges from 0 (worst possible life) to 10 (best possible life). Population means in low-income countries typically fall between 4 and 6. A treatment coefficient of 0.3–0.5 ladder steps is considered a meaningful programme effect based on the literature; smaller effects may be substantively real but are hard to detect without large samples.alphaoutput from the self-efficacy scale should ideally exceed 0.70 (Cronbach’s alpha ≥ 0.70 is the conventional minimum for internal consistency). Values below 0.60 suggest the items are not reliably measuring the same construct and warrant review of the translation or scale adaptation.- A negative treatment coefficient on
phq9_scoremeans the treatment arm has lower depression scores (an improvement). A positive treatment coefficient oncantril_laddermeans the treatment arm reports higher life satisfaction. Report effect sizes alongside raw coefficients — a one standard deviation change in the comparison population is a useful benchmark.
References
Banerjee, A., Duflo, E., Goldberg, N., Karlan, D., Osei, R., Parienté, W., Shapiro, J., Thuysbaert, B., & Udry, C. (2015). A multifaceted program causes lasting progress for the very poor: Evidence from six countries. Science, 348(6236). https://doi.org/10.1126/science.1260799
Blattman, C., Jamison, J. C., & Sheridan, M. (2017). Reducing crime and violence: Experimental evidence from cognitive behavioral therapy in Liberia. American Economic Review, 107(4), 1165–1206. https://doi.org/10.1257/aer.20150503
Duckworth, A. L., Peterson, C., Matthews, M. D., & Kelly, D. R. (2007). Grit: Perseverance and passion for long-term goals. Journal of Personality and Social Psychology, 92(6), 1087–1101. https://doi.org/10.1037/0022-3514.92.6.1087
Haushofer, J., & Shapiro, J. (2016). The short-term impact of unconditional cash transfers to the poor: Experimental evidence from Kenya. Quarterly Journal of Economics, 131(4), 1973–2042. https://doi.org/10.1093/qje/qjw025
Heckman, J. J., Stixrud, J., & Urzua, S. (2006). The effects of cognitive and noncognitive abilities on labor market outcomes and social behavior. Journal of Labor Economics, 24(3), 411–482. https://doi.org/10.1086/504455
Schwarzer, R., & Jerusalem, M. (1995). Generalized self-efficacy scale. In J. Weinman, S. Wright, & M. Johnston (Eds.), Measures in Health Psychology: A User’s Portfolio (pp. 35–37). NFER-Nelson.