NOTE-1530: Murphy et al. — The BPNSFS probably does not validly measure need frustration

Contribution

It makes the case that the Basic Psychological Need Satisfaction and Frustration Scales (Chen et al. 2015), “the dominant self-report measure in the BPNT domain”, cannot carry the two-dimensional theory of need satisfaction and need frustration (Vansteenkiste & Ryan 2013) that they are used to test. It gives an item-content argument, a factor-analytic replication of Chen’s own 2013 dissertation finding in new data, and a test of whether the scales’ different correlates track the keying of the external measures. A second, separate finding: the two autonomy scales may measure different things.

Key insight

When every “frustration” item is negatively keyed and every “satisfaction” item positively keyed, a two-factor split and a difference in correlates are exactly what keying alone would produce, so neither can show that the two scales measure two constructs (Kam & Meyer 2015). The one thing the structure can show is which grouping is stronger. Theory requires the needs (competence, relatedness, autonomy) to be the higher-order factors. In these data the keying directions are. “Item-keying direction was an overwhelming source of the covariance in the BPNSFS item pool, more so than the three basic psychological needs dimensions themselves” (Discussion, Issue A).

Assumptions

• The construct being tested is Vansteenkiste and Ryan’s (2013). Need frustration is “an asymmetric configuration of (a) low need satisfaction alongside (b) strong, active, direct thwarting by other individuals or by the social context itself” (introduction). The plant analogy: lack of rain versus salting the soil. Items without thwarting content are judged against that definition.
• The needs should be the highest-order factors if the scale measures what BPNT says, since “frustration of any particular need is conceptualized, by definition, as inherently covarying with need satisfaction of that need” (introduction, after Campbell & Fiske 1959).
• Keying and valence coincide. In the BPNSFS, positive keying is also positive valence, so keying factors may carry defensiveness, self-esteem or social desirability (DiStefano & Motl 2009; Quilty et al. 2007).
• Pooling. Items were divided by the scale maximum (5 or 7) and treated as continuous; the eight samples were pooled, with Sample 2’s primed half as a cluster (Method). Seven samples are university students, one an online community sample; all English-language.

Key results

The items (introduction)

• None of the three frustration scales has “substantial active thwarting content”. Two Competence Frustration items are reverse-worded twins of satisfaction items: “I have serious doubts about whether I can do things well” against “I feel confident that I can do things well”. The other two concern failures (“I feel like a failure because of the mistakes I make”).
• Chen (2013), the dissertation behind Chen et al. (2015), “tentatively concluded that the item pool did not support the creation of separate frustration scales”, found two large keying factors in an EFA, and modelled the negative items as a method. The 2015 paper presented six scales from the same pool.
• Footnote 2 contrasts the Psychological Need Thwarting Scale (Bartholomew et al. 2011, LIT-1954), which “measures being thwarted far more explicitly”, e.g. “I feel forced to follow decisions made for me”.

Issue A: keying (Results)

• External correlates follow the external items’ keying (all items scored toward well-being). Self-compassion: Satisfaction correlates .35 with the positive SCS items and Frustration .09 (Steiger’s z = 4.22); Frustration .67 with the negative items, Satisfaction .34 (z = 6.50). La Guardia need-satisfaction scale: positive items .78 vs .58 (z = 7.80); negative items .75 vs .45 (z = 11.16). Perceived stress (two samples): positive items .48 vs .53 (z = −.66), negative items .65 vs .46 (z = 3.58). Depression (CESD): positive items .59 vs .56 (z = .49), negative items .62 vs .53 (z = 1.63). Self-esteem (RSE): positive .59 vs .57 (z = .33), negative .67 vs .57 (z = 1.90). So Frustration’s excess with negative items holds in all five measures, clearly in three; Satisfaction’s excess with positive items holds in two of five. The paper’s word for it is “though not consistently”.
• EFA (Table 1, N = 3,692). Two factors: one of all positive items, one of all negative items. Three factors: the first two still pure keying, with some relatedness items on the third. Four factors: a frustration factor (competence and relatedness), a satisfaction factor (autonomy and relatedness), a relatedness factor of both keyings, and an autonomy frustration factor. Parallel analysis suggested six factors, MAP four.
• Six-factor EFA intercorrelations. Among the frustration factors .64 to .70; among the satisfaction factors .66 to .75; between satisfaction and frustration of the same need .17 (competence), .34 (relatedness), .38 (autonomy).
• Higher-order CFAs. Six first-order factors under three need factors: RMSEA .064, CFI .924, TLI .915, BIC −28,734, with the higher-order correlations .94 to 1.11, “exceed[ing] 1 in most cases”. Under two keying factors: RMSEA .052, CFI .950, TLI .944, BIC −29,631.

Issue B: the autonomy scales (Results, Table 3)

• Three substantive factors plus two keying factors fit well (RMSEA .033, CFI .981, TLI .977). In Table 3 the four Autonomy Satisfaction items load .49, .34, .36, .42 on Autonomy, below the .50 threshold, and the four Autonomy Frustration items load .76, .62, .66, .75. Every relatedness and competence item but one (.48) loads above .50.
• Four substantive factors (autonomy split in two) fit “nearly equivalently” (RMSEA .033, CFI .982, TLI .978); the two autonomy factors correlate .82.
• Correlates. Autonomy Satisfaction correlates .39 with the IAF’s Authorship/Self-Congruence (Sample 3) and .52 with presence of meaning (Sample 1); Autonomy Frustration −.08 and .15.

Conclusions (Discussion)

• “The BPNSFS probably does not validly measure the dual-dimension theory of need frustration and need satisfaction”, and “many published findings regarding well-being that rely on the BPNSFS may be at least partly driven by methodological artifact”.
• Chen et al. (2015)‘s nomological differences are explained by keying: need satisfaction tracked the Satisfaction with Life Scale (no negative items) and vitality, frustration the CESD (mostly negative items).
• A three-factor model (with method factors) is “better justified than a six-factor model”; a four-factor one is suggested and not asserted.

Claims

id	claim	strength	support
C1	The BPNSFS frustration items lack the active-thwarting content the two-dimensional theory defines frustration by	strong	the items themselves, quoted (introduction); conceded by Kermavnar et al., whose wording follows Chen (2013)
C2	In the BPNSFS item pool, keying direction is a stronger source of covariance than need domain	strong	EFAs with 2–4 factors, six-factor intercorrelations, two higher-order CFAs, pooled N = 3,692
C3	The satisfaction–frustration split is “likely primarily” a keying artefact	moderate	C1 and C2, plus external correlates sorted by keying; the authors say this cannot be settled dispositively because keying and construct are confounded
C4	The BPNSFS frustration scales’ stronger link to ill-being is explained by keying	moderate	Frustration’s excess with negatively keyed external items in five measures, significant in three; offered as a “rebuttable presumption”
C5	Autonomy Satisfaction and Autonomy Frustration are not commensurable	weak	external correlates in two samples, mostly positively keyed; the four-factor model did not improve fit and its autonomy factors correlate .82

Concepts

• Item keying — whether agreeing with an item indicates more (positive) or less (negative) of the construct; here also the item’s valence.
• Need frustration (as the paper uses it) — Vansteenkiste and Ryan’s: low satisfaction plus active thwarting by others or the context; not a synonym for low satisfaction, which it was before 2013.
• Method factor (“artifactor”) — a factor shared by items because of how they are worded rather than what they are about.
• Rebuttable presumption — if a scale’s two keyings each correlate preferentially with external items of the same valence, keying is presumed to drive the difference until shown otherwise.

Connections

• Chen (2013) and Chen et al. (2015). The dissertation and the published scale; the paper’s arguments “are corroborations of Chen’s (2013) insightful conclusions”, in new samples. Chen et al. is held as LIT-2000, unread.
• Kam and Meyer (2015), on the Rosenberg scale: the method of testing whether keying drives a nomological difference.
• Bartholomew et al. (2011), LIT-1954, the thwarting scale, as the contrast: explicitly thwarting items.
• Vansteenkiste, Ryan and Soenens (2020), LIT-561, which “promoted” the BPNSFS as valid for both constructs.
• Brett Murphy, the first author, was a named reviewer of Kermavnar et al. 2024 (LIT-1988), which adopts this paper’s conclusion.

Bearing on the record

• THEORY-052 (need frustration as transdiagnostic). Its “What was actually shown” opens “Measured directly (BPNSFS and similar scales), frustration predicted burnout, eating symptoms and negative affect beyond low satisfaction”. This paper says that, for the BPNSFS, “beyond low satisfaction” is what keying alone would produce, and that the satisfaction and frustration factors of the same need correlate only .17 to .38 after the keying is pulled apart. So the separation of frustration from low satisfaction, as measured by the BPNSFS, is not evidence that frustration is a separate dimension. Campbell et al. 2018, its one longitudinal study, measured need frustration with exactly these twelve items (NOTE-1533). The objection is distinct from the shared-content one THEORY-052 already notes, as its 2026-11-02 note says, and both hit the same evidence. Its promote_when already asks for a measure that is not a distress-sharing self-report; it should also exclude a frustration score whose separation from satisfaction is a keying split.
• QUESTION-131 (structural frustration and operative thwarting). The paper is a measurement-level case of the question’s distinction. The construct SDT names frustration is defined by an identified thwarting agent, which is the question’s “operative thwarting”. The BPNSFS items carry no agent and measure, on this reading, low satisfaction: the question’s middle row, “non-attainment”, which the question says is neither. The thwarting scale (footnote 2) is the instrument that names an agent. So the psychology already separates the three rows, and its most used scale collapses the third into the second.
• QUESTION-369 (what would show a need is a need). The frustration test (frustration predicts ill-being beyond low satisfaction) is one of SDT’s routes to need-hood. If the BPNSFS frustration score is a keyed satisfaction score, a candidate need’s frustration “adding variance” can be a fact about keying; and its correlates sort by the valence of the outcome measure, so the test is partly the well-being measure again. This is the interdefinition NOTE-439 records, at the level of the instrument.
• CLAIM-368 cites THEORY-052 as a parallel only (“keeps need frustration apart from low satisfaction”). That sentence is true of the theory and, on this reading, not established of the BPNSFS measurements. The claim’s own case does not rest on it.
• THEORYs filed from this reading: THEORY-1316 (Active), what the item content and keying structure show; THEORY-1311 (Proposed), that the frustration scales are therefore keyed low satisfaction.
• No instruction for machine-learning practice.

Limitations

• Not dispositive, by the authors’ account. Keying and construct are fully confounded in the BPNSFS, so no factor analysis can separate them; the proposed fix, reverse-worded twins of all 24 items, was not run.
• Secondary data. None of the eight samples was collected to test the BPNSFS, and only a few had the external measures; the self-compassion, La Guardia and meaning analyses rest on one sample each.
• The keying test is mixed. Satisfaction’s preference for positive external items appeared in two of five measures; for depression and self- esteem it did not. The depression and self-esteem negative-item differences (z = 1.63, 1.90) are marginal at best.
• Narrow samples: English-language, mostly undergraduates, three of the BPNSFS’s 50-odd versions.
• Not preregistered (Method).
• An internal inconsistency on the autonomy items. Results says of the three-factor model that “none of the Autonomy Satisfaction items loaded above .50”, which Table 3 bears out (.34–.49). The Discussion says instead that “none of the Autonomy Frustration items loaded well on an Autonomy factor”, which Table 3 contradicts (.62–.76). Of the four-factor model, Results says “two of the four Autonomy Frustration items still did not load above .5”, but in Table 3 the two items below .50 are Autonomy Satisfaction (.44, .46); every Autonomy Frustration item loads .62 or more. The introduction’s report of Chen (2013) has the frustration items loading weakly. So the paper’s own data have the satisfaction items, not the frustration items, as the weak ones, and the argument of Issue B, that the frustration items lack self-endorsement content, is supported by the external correlates and not by these loadings. This is a reading of the manuscript’s table; the published version was not checked.

Open questions

• With reverse-worded twins of every item, do satisfaction and frustration still separate? That is the test the authors propose.
• Does a frustration scale built from thwarting content (the PNTS, or the GNSF Kermavnar et al. single out) separate from satisfaction when its keying is balanced, and does it predict ill-being beyond low satisfaction?
• How much of THEORY-052’s evidence used the BPNSFS? Campbell et al. 2018 did.