Acute Effect of Alcohol on Working Memory Updating

Abstract

Aims:

To examine the acute effects of alcohol on Working Memory (WM) Updating, including potential variation across the ascending limb (AL) and descending limb (DL) of the blood alcohol concentration (BAC) time course.

Design:

A two-session experiment in which participants were randomly assigned to one of three beverage conditions (alcohol [males: 0.80 g/kg; females: 0.72 g/kg], active placebo [0.04 g/kg], or non-alcohol control [tonic]) and one of two BAC limb testing conditions (AL and DL, or DL-Only) for the second session, yielding a 3 (beverage) x 2 (timepoints tested) x 3 (timepoint) mixed factorial design with repeated measures on the latter factor. One of the repeated assessments is “missing by design” in the DL-Only condition.

Setting:

A psychology laboratory at the University of Missouri campus in Columbia, MO, USA.

Participants:

Two hundred thirty-one community-dwelling young adults (51% female; aged 21–34 years) recruited from Columbia, MO, USA, and tested between 2011 and 2013.

Measurements:

Latent WM Updating performance as indexed by shared variance in accuracy on three WM Updating tasks (Letter Memory, Keep Track, Spatial 2-Back) at three timepoints.

Findings:

Multigroup modeling of latent WM Updating indicated that performance among participants who consumed placebo or control beverages improved during the second session at timepoints corresponding to AL (Δ from baseline in latent M±SE: .+05±.01, p<.001) and DL (+.08±.01, p<.001). Alcohol consumption did not impair WM Updating (Δ from baseline in latent M±SE, at AL: +.01±.01, p=.56; at DL: +.05±.01, p<.001), but attenuated performance improvements (equality of latent Ms across beverage groups at AL or DL: Δχ² (1)≥7.53, p<.01).

Conclusions:

Acute alcohol-induced impairment in Working Memory Updating may be limited, but dampening of practice effects by alcohol could interfere with the completion of novel, unpracticed tasks.

Numerous theoretical models^1–4 posit that alcohol-related hazards arise from alcohol-induced impairment in executive functioning (EF). EF refers to higher-order cognitive processes that support control over thoughts and actions during goal-directed behavior,⁵ which can be organized into general and specific components.⁶ A subcomponent of EF, Working Memory (WM) Updating, reflects the ability to maintain and manipulate existing information in WM while dynamically replacing or updating other information in WM. Broadly, acute impairment of EF is believed to underlie impaired control over drinking and negative consequences.^7–10 Intoxicated individuals may experience several difficulties due to WM Updating impairment, including trouble keeping track of numbers of drinks consumed and rendering accurate mental maps of locations, conversations, and changing situations. In this way, intoxicated individuals may not maintain awareness of long-term goals and/or shield them against competing short-term goals and temptations.¹¹

Evidence for acute effects of alcohol on WM Updating is mixed. A recent systematic review (k=13) concluded that performance on auditory/speech-related WM tasks was reliably impaired by alcohol even at moderate doses, whereas performance on visuospatial WM tasks was commonly spared even at higher doses.¹² Others have questioned that conclusion by showing that performance on auditory/speech-related WM tasks can be unaffected at low to moderate doses,^13–15 and performance on visuospatial WM tasks can be impaired at moderate to high doses^16,17 and in daily life.¹⁸

Limitations of Previous Studies

Previous studies of alcohol and WM have been limited in at least five ways. First, most studies have utilized a single WM Updating task,^19–22 making it difficult to generalize beyond a specific task to the broader construct due to “task impurity.”^6,23 Laboratory tasks often suffer from low reliability,²⁴ so scores on individual WM Updating tasks can reflect the influence of other EF facets, non-EF-related cognitive processes, and measurement error.⁶ Thus, divergent effects reported in the literature may arise from task heterogeneity and unreliability. Latent variable models circumvent this issue by capturing shared variance among construct-relevant tasks and accounting for measurement error. Second, most studies use small samples (N ≈ 25),¹² which are both underpowered and more likely to produce false positive findings. Third, acute alcohol effects are tested almost entirely against a single control condition, either placebo-alcohol^25,26 or a no-alcohol beverage.^22,27 Including placebo and no-alcohol control conditions allows testing whether acute effects are driven by expectancy, pharmacology, or both.²⁸

Fourth, most existing studies examined alcohol effects while blood alcohol concentration (BAC) is rising or at its peak,^29–31 but alcohol’s effect on WM Updating may differ within a drinking episode. For instance, WM Updating may improve on the descending relative to ascending limb of the BAC curve due to acute tolerance.^32,33 Subjective intoxication appears reliably affected by acute tolerance, but acute tolerance effects on impaired performance are neither reliable nor uniform across cognitive-behavioral domains.^34,35 Few studies have examined acute tolerance during WM Updating,^15,22,36 although this phenomenon could cause acute alcohol-related hazards. Fifth, previous work has not considered practice effects. Repeated exposure to EF tasks improves task performance, which may indicate decreased difficulty or novelty, and in some cases, the development of better strategies.^37,38 Acute alcohol effects could manifest in impaired WM Updating performance relative to baseline, or in the blunting or elimination of practice effects.

Current Study

Here, these limitations were addressed with several design features. We used a relatively large community sample (N = 231) of young adults, representing the developmental period in which heavy drinking, alcohol-related problems, and alcohol use disorder (AUD) are most prevalent.^39–42 Three widely-used WM Updating tasks^43–50 were completed during two laboratory sessions separated by 1–3 weeks. To circumvent task impurity, we derived latent variables from performance across the three tasks. The two-session feature allowed for identifying between-subject differences in baseline ability and within-subject changes in performance between and within sessions. We also compared performance under alcohol to active placebo and no-alcohol control beverage conditions to disentangle alcohol’s expected and pharmacological effects. Finally, WM Updating was assessed either once or twice after beverage consumption (either while BAC was descending, or while it was both ascending and descending). This “missing by design”⁵¹ feature permits examination of whether any differences in alcohol’s effects during ascending and descending BAC reflect acute tolerance or practice effects.

The study had three main aims: (1) test the extent to which alcohol acutely impairs WM Updating; (2) examine whether acute effects of alcohol on WM Updating differ during ascending versus descending BAC; and (3) determine whether any such differences reflect acute tolerance or practice effects.

Method

University of Missouri Institutional Review Board approved all procedures. Analyses were planned prior to data collection in the grant application that funded the study (P60 AA011998 5979). However, analyses were not formally pre-registered, so results should be considered exploratory.

Participants

Two hundred thirty-one healthy young adults (51% female; 86% Caucasian) aged 21–34 years (M = 23.14, SD = 2.74) who reported regular alcohol use (2–25 unitsⁱ per week and at least one heavy use occasionⁱⁱ over the past year) and no contraindications to alcohol administration (see Supplemental Materials) were recruited from Columbia, MO, as in our two previous, related studies.^52,53 Table 1 presents demographics and recent alcohol use for individuals randomly assigned to each cell of the experiment.

Table 1.

Characteristics of the Sample (N = 231)

	Alcohol (n = 85)		Placebo (n = 71)		Control (n = 75)
	A/D (n = 46)	D–Only (n = 39)	A/D (n = 36)	D–Only (n = 35)	A/D (n = 38)	D–Only (n = 37)	χ 2	p
Age in years, Med (IQR)	22.17 (3.81)	22.19 (1.98)	21.90 (1.65)	22.19 (1.61)	21.94 (1.70)	21.96 (1.96)	2.62	.758
Sex, n Female (%)	23 (50)	20 (51)	18 (50)	17 (48)	19 (50)	19 (51)	0.08	.999
Ethnicity, n Hispanic (%)	1 (2)	3 (8)	3 (8)	0 (0)	2 (5)	4 (11)	5.81	.325
Race, n (%)							33.09	.130
American Indian or Alaska Native	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	1 (3)
Asian	0 (0)	3 (8)	0 (0)	0 (0)	0 (0)	0 (0)
Black or African American	5 (11)	0 (0)	1 (3)	5 (14)	3 (8)	0 (0)
Native Hawaiian or Pacific Islander	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
White	39 (85)	32 (82)	33 (92)	28 (80)	33 (87)	34 (92)
No information	2 (4)	4 (10)	2 (5)	2 (6)	2 (5)	2 (5)
Alcohol Use, Med (IQR)
Times Used in past 3 months	19.29 (31.07)	19.29 (12.86)	19.29 (37.50)	19.29 (37.50)	19.29 (0.00)	19.29 (25.71)	3.06	.691
Typical Units in past 3 months	5.00 (3.00)	5.00 (2.00)	5.00 (2.50)	3.00 (2.50)	5.00 (3.00)	3.00 (2.00)	3.49	.625
Times Used in past 30 days	6.43 (0.00)	6.43 (4.29)	6.43 (3.93)	6.43 (12.5)	6.43 (8.57)	6.43 (0.00)	1.55	.907
Typical Units in past 30 days	4.00 (3.50)	4.00 (2.00)	4.00 (3.00)	4.00 (3.00)	5.00 (3.00)	3.00 (3.00)	3.88	.567
Times 5+ Units in past 30 days	2.50 (6.18)	2.50 (5.43)	1.00 (3.48)	1.00 (5.43)	2.50 (5.43)	1.00 (6.43)	7.37	.194
Times 12+ Units in past 30 days	0.00 (0.00)	0.00 (1.00)	0.00 (0.00)	0.00 (0.50)	0.00 (0.00)	0.00 (1.00)	1.15	.950
Max Units Used in past 30 days	7.50 (5.00)	8.00 (6.00)	7.00 (4.87)	7.00 (5.50)	8.00 (4.00)	5.00 (8.00)	5.63	.344
Max Units Used in lifetime	15.00 (10.00)	16.00 (9.50)	14.50 (11.25)	13.00 (6.50)	12.25 (8.50)	15.00 (8.00)	5.14	.399

Note. A/D refers to the subset of participants assigned to complete WM Updating tasks on both the ascending and descending limbs of the BAC time course or corresponding timepoints. D-Only refers to the subset of participants assigned to complete WM Updating tasks only on the descending limb of the BAC timecourse or corresponding timepoints. Med = median. IQR = inter-quartile range. Med and IQR are given for distributions of age and alcohol use because these were not normally distributed. With respect to alcohol use, 1 unit here refers to U.S. standard serving equivalents (14 g ethanol). See Supplemental Materials for alcohol use question language, response options, and scaling. The similarity of the distributions of age and alcohol use across the six experimental cells to which participants were randomized was verified using Kruskal-Wallis χ² tests with 5 df whereas the similarity of the distributions of sex, ethnicity, and race categories was verified using Pearson χ² tests with 5, 5, and 25 df, respectively.

Measures

Breath Alcohol Concentration (BrAC)

BrAC was measured using an Alco-Sensor IV (Intoximeters, St. Louis, MO) as g/210 L exhaled air, which is equivalent to g/dL whole blood, and is reported here as g%. BrACs are a reliable^54,55 proxy for expected post-absorption BACs.^56,57 In the alcohol and placebo conditions, BrAC was measured every 15–30 min after the end of the beverage administration period (24 min consumption, 5 min absorption) with care taken not to interrupt task set completion. BrACs were not shared with participants.

Placebo Manipulation Check Items

At the end of the experimental session, participants assigned to the alcohol and placebo conditions were asked to rate their subjective intoxication (0–4, “not at all” to “a lot”) after beverage consumption as well as during the AL and DL procedures. Additionally, they were asked to indicate “the number of standard drinks you think would be equivalent to what you drank in the study today,” using integers 0–20.

WM Updating Tasks

WM Updating was measured with three widely-used^43–50 tasks: Keep Track,^6,44 Letter Memory,^6,43 and Spatial 2-Back,^45,46 each of which is described briefly below (see Supplemental Materials for detailed descriptions). Task scores reflected the proportion of correct responses. Internal reliability varied across tasks, but was reasonable in all tasks (see Table 2). Test-rest reliability also was reasonable (see Table 3). Moreover, as in previous reports,^6,46,48,58 tasks were moderately correlated with one another at each assessment (see Table 4), supporting the notion that they measure the same underlying construct.

Table 2.

Internal consistency reliability of raw accuracy scores for each task at every assessment

	ICC			r split-half
Tasks	Baseline	AL	DL	Baseline	AL	DL
KT	.50	.50	.61	.27	.37	.40
LM	.57	.68	.72	.47	.547	.60
SNB	.88	.87	.89	.82	.82	.83

Note. KT = Keep Track task; LM = Letter Memory task; SNB = Spatial 2-Back task. Baseline represents assessment at the baseline session. AL stands for ascending limb (or corresponding timepoint) assessment in the experimental session. DL stands for descending limb for ascending limb (or corresponding timepoint) assessment in the experimental session. ICC stands for intraclass correlation coefficient. Specifically, ICC (3, k)⁷⁹ was computed, using the k trials in each task as the k raters. Split-half correlation (r_split-half) was computed as Pearson’s r for the average score across even- vs. odd-numbered trials in each task. The number of participants contributing pairwise complete data was: for KT, Baseline n = 231, AL n = 120, DL n = 228; for LM, Baseline n = 231, AL n = 119, DL n = 227; and for SNB, Baseline n = 230, AL n = 119, DL n = 230. All ICC and r_split-half were significantly different from 0 at p < .001.

Table 3.

Test-retest reliability of raw and transformed accuracy scores for each task across assessment

		ICC								r test-retest
		Baseline vs. AL		Baseline vs. DL		AL vs. DL		Baseline vs. AL vs. DL		Baseline vs. AL		Baseline vs. DL		AL vs. DL
Grp.	Tasks	Raw	Transf	Raw	Transf	Raw	Transf	Raw	Transf	Raw	Transf	Raw	Transf	Raw	Transf
All	KT	.73	.74	.61	.62	.77	.77	.8	.8	.57	.58	.44	.45	.63	.63
	LM	.67	.71	.66	.66	.78	.78	.77	.79	.51	.55	.49	.5	.65	.64
	SNB	.81	.82	.78	.76	.87	.87	.87	.87	.69	.71	.64	.64	.78	.77
Cntrl	KT	.76	.78	.55	.59	.84	.83	.81	.82	.61	.65	.38	.41	.73	.71
	LM	.67	.67	.57	.59	.75	.77	.76	.78	.5	.51	.4	.42	.66	.66
	SNB	.83	.82	.72	.69	.91	.88	.89	.87	.72	.74	.58	.56	.82	.78

Note. Grp = Groups Included. Cntrl = Control Beverage Only. Transf = Transformed. KT = Keep Track task; LM = Letter Memory task; SNB = Spatial 2-Back task. Baseline represents assessment at the baseline session. AL stands for ascending limb (or corresponding timepoint) assessment in the experimental session. DL stands for descending limb for ascending limb (or corresponding timepoint) assessment in the experimental session. ICC stands for intraclass correlation coefficient. Specifically, ICC (3, k)⁷⁹ was computed, using k = 2 or 3 tests, as the k raters. Test-retest correlations (r_test-retest) were computed as Pearson’s r. For ease of comparison with the extant literature, correlations are presented for both raw and transformed (angularized and winsorized; see Analytic Strategy) task score. For Groups Included = All, the number of participants contributing complete data was: for KT, Baseline vs. AL n = 120, Baseline vs. DL n = 228, AL vs. DL n = 120, Baseline vs. AL vs. DL n = 120; for LM, Baseline vs. AL n = 119, Baseline vs. DL n = 227, AL vs. DL n = 119, Baseline vs. AL vs. DL n = 119; and for SNB, Baseline vs. AL n = 119, Baseline vs. DL n = 228, AL vs. DL n = 119, Baseline vs. AL vs. DL n = 119. For Groups Included = Control Only, the number of participants contributing complete data was: for KT, Baseline vs. AL n = 38, Baseline vs. DL n = 73, AL vs. DL n = 38, Baseline vs. AL vs. DL n = 38; for LM, Baseline vs. AL n = 36, Baseline vs. DL n = 74, AL vs. DL n = 38, Baseline vs. AL vs. DL n = 38; and for SNB, Baseline vs. AL n = 36, Baseline vs. DL n = 72, AL vs. DL n = 38, Baseline vs. AL vs. DL n = 38. All ICC and r_test-retest were significantly different from 0 at p < .001.

Table 4.

Intercorrelation of raw and transformed task accuracy scores at every assessment

	Baseline		AL		DL
Tasks	Raw	Transformed	Raw	Transformed	Raw	Transformed
KT vs. LM	.18**	.22***	.37***	.41***	.28***	.34***
KT vs. SNB	.11	.14*	.38***	.37**	.25***	.26***
LM vs. SNB	.26***	.28***	.24**	.28**	.19**	.24***

Note. KT = Keep Track task; LM = Letter Memory task; SNB = Spatial 2-Back task. Baseline represents assessment at the baseline session. AL = ascending limb (or corresponding timepoint) assessment in the experimental session; DL = descending limb (or corresponding timepoint) assessment in the experimental session. Correlations were computed as Pearson’s r. For ease of comparison with the extant literature, correlations are presented for both raw and transformed (angularized and winsorized; see Analytic Strategy) task scores. The number of participants contributing pairwise complete data was: for Baseline, KT vs. LM n = 231, KT vs. SNB n = 229, LM vs. SNB n = 229; for AL, KT vs. LM n = 119, KT vs. SNB n = 119, LM vs. SNB n = 118; and for DL, KT vs. LM n = 226, KT vs. SNB n = 227, LM vs. SNB n = 226.

^*p < .05

^**p < .01

^***p < .001

Keep Track Task

After seeing a sequence of 15–25 words, participants were asked to recall and repeat aloud the most recently presented exemplar from 3–5 distinct categories (e.g., animals, countries).

Letter Memory Task

After seeing a sequence of 9–13 letters, participants were asked to recall and repeat aloud the four most recent letters in the order of presentation.

Spatial 2-Back Task

Across a sequence of visual “flashes” at different locations on a monitor, participants were asked to indicate via button press whether the location of the most recent “flash” matched or did not match the location two “flashes” back in the sequence.

Procedure

Figure 1 shows the overall design, including the baseline and experimental sessions, key events within each session, and randomization of participants to one of six conditions for the experimental session.

Figure 1. Schematic overview of experimental design.

Note. Bev. Admin. Stands for beverage administration. Baseline sessions (3–4 hr) were held between 9am and 1pm. Experimental sessions (4 hr) were held between 12pm and 5pm. Sessions were held 1–3 weeks apart. Baseline Session: after BrAC testing to confirm sobriety, participants provided informed consent and completed the three WM Updating tasks (Letter Memory [LM], Keep Track [KT], and Spatial N-Back [SNB]; squares labeled ‘1’, ‘2’, ‘3’) and six other EF tasks (rectangle labeled ‘…’; not reported here). Tasks were completed in a fixed order counterbalanced across participants. Experimental Session: After BrAC testing to confirm sobriety, participants provided informed consent again and were asked to void the bladder, during which time, female participants also self-administered a hormone-based (urine) pregnancy test in a private restroom (all tested negative). Participants were then administered a beverage and completed the three WM Updating tasks (LM, KT, SNB; squares labeled ‘1’, ‘2’, ‘3’) and two other EF tasks (squares labeled ‘…’; not reported here) once or twice during the session according to randomly assigned experimental conditions. Participants were randomly assigned to one of six experimental cells resulting from fully crossing a 3-level beverage condition with a 2-level assessment condition. Beverage conditions were alcohol (0.80 g/kg for males, 0.72 g/kg for females), active placebo alcohol (0.04 g/kg), and no-alcohol control (tonic only). Participants assigned to the control condition were told that the beverage contained “no alcohol” whereas participants in the alcohol and placebo conditions were told that the beverage contained “a moderate amount of alcohol.” Assessment conditions were A/D (tasks completed during both ascending and descending limbs of the BAC timecourse or corresponding timepoints) and D-Only (tasks completed only during descending limb of the BAC timecourse or corresponding timepoints). Following beverage administration, participants assigned to the A/D condition completed the tasks when BrAC had risen to at least .065 g% (or corresponding timepoints) whereas participants assigned to the D-Only condition watched episodes of a popular sitcom (The Office [U.S.]). After peak BrAC (≈0.085 g%), participants in both the A/D and D-Only conditions completed the tasks (in reverse order) when BrAC had fallen to at least 0.075 g% (or corresponding timepoints).

Analytic Strategy

Following previous work,^45,46 task scores were winsorized at ±3 SD from the mean (to reduce the influence of extreme values but retain their ordinal positions) and then subjected to angular transformation (arcsine of the square root) to normalize their distributions.⁵⁹

Latent variable models were estimated in Mplus version 7.3⁶⁰ using the robust maximum likelihood estimator and full-information maximum likelihood,⁶¹ which can handle the “planned missingness” feature of this study (i.e., D-Only condition; see Figure 1) as well as data missing at random.ⁱⁱⁱ Adequacy of model fit was based on the following guidelines suggested in the literature: Comparative Fit Index (CFI) and Tucker-Lewis index (TLI) > .95 for reasonably good fit; and root mean square error of approximation (RMSEA) ≤ .08 for reasonable fit and ≤ .05 for close fit. We also report model fit χ² significance tests but did not rely on them because they are highly sensitive to N.⁶²

To examine groups’ latent WM Updating performance across testing occasions, we specified a series of multigroup models (Figure 2). We established strict measurement and temporal invariance on the structural model (equivalent task loadings, intercepts, and residual variances on latent WM Updating factors across groups and occasions; see Table S5 for invariance analyses), which is critical to rule out the possibility that observed differences in latent mean values across groups and occasions are due to measurement differences rather than substantive change.⁶³ We then tested whether latent means could be constrained to equality across groups and occasions (see Table S5 for invariance analyses). Here, models were compared using the ΔCFI (|.002|) and ΔRMSEA (|.015|); we gave preference to ΔCFI because it is more stringent and widely accepted.^64,65 We report Satorra-Bentler Δχ ² tests but did not rely on them because they are overly sensitive to peripheral factors (e.g., N).⁶⁵

Figure 2. Conceptual Diagram of the Latent Variable Model of WM Updating Performance.

Note. KT = Keep Track Task; LM = Letter Memory Task; SNB = Spatial N (2)-Back Task. Baseline refers to the baseline session. Ascending Limb and Descending Limb refer to different periods within the experimental session (completed 1–3 weeks after baseline session under one of six experimental conditions). See Figure 1 for procedural details, and Analytic Strategy for multigroup modeling details. This figure depicts a conceptual diagram reflecting the multigroup models we tested. We examined pharmacological effects of alcohol by testing differences in latent WM Updating means across “Alcohol” groups (0 = No Alcohol [placebo/control], 1 = Alcohol [alcohol]) and alcohol expectancy effects with “Expectancy” groups (first, defined as: 0 = No Expectancy [control], 1 = Expectancy [alcohol/placebo]. To disentangle “pure” expectancy effects from pharmacological effects, we also tested “Pure Expectancy” groups (0 = No Expectancy [control], 1 = Expectancy [placebo]). Thus, multigroup models comprised two groups (No Alcohol vs. Alcohol, No Expectancy vs. Expectancy), each with three lower-order latent factors (i.e., one per testing occasion), one higher-order (cross-occasion) factor onto which each lower-order factor loaded, and residual covariances among tasks across occasions and within groups. The WM Updating latent factor from the Baseline testing occasion of either the No Alcohol or No Expectancy group served as the reference factor. The most invariant multi-group models, which equated task factor loadings, intercepts, and residual variances across groups and occasions, fit well (Alcohol/No Alcohol: X² = 83, df = 59, CFI = .961, TLI = .953, RMSEA = .059; Expectancy/No Expectancy: X² = 82, df = 59, CFI = .963, TLI = .954, RMSEA = .058; see Table S5 for invariance analyses and Table S6 for final model parameters).

Results

BrAC at start of AL and DL procedures

Table 5 presents BrAC Ms and SDs. BrACs were analyzed using a 2 (Task Completion Group: A/D, D-Only) x 2 (BAC Curve Limb: AL, DL) mixed factorial ANOVA with repeated measures on the latter factor. There was a significant main effect of BAC Curve Limb, F(1, 83)=22.26, p<.001, such that BrACs were lower at DL than AL, t(84)=4.96, p<.001. Neither the main effect of Task Completion Group, F(1, 83)=1.52, p=.221, nor the Task Completion Group x BAC Curve Limb interaction effect, F(1, 83)=2.36, p=.128, were significant.

Table 5.

BrAC (g%) at start of AL and DL procedures for participants assigned to Alcohol condition

	AL		DL
By WM Updating Task Completion condition	M	SD	M	SD
A/D	.075	.014	.067	.008
D-Only	.071	.013	.067	.007
Collapsing WM Updating Task Completion condition	.073	.013	.067	.008

Note. AL = ascending limb and DL = descending limb of the breath alcohol concentration (BrAC) curve. A/D refers to 46 participants who completed the WM Updating tasks on both the AL and DL. D-Only refers to 39 participants who completed the WM Updating tasks only on the DL. Although D-Only participants completed WM Updating tasks only on the DL, their BrAC values were recorded on both the AL and DL.

Placebo Manipulation Check

Almost all placebo participants (69/71) estimated consuming a non-zero number of alcoholic drinks during the study, but those numbers, as well as retrospective subjective intoxication ratings, were lower among placebo compared to alcohol participants (see Table 6).

Table 6.

Post-Experiment Expectancy Manipulation Checks

	Alcohol							Placebo
	M	SD	Med	IQR	R	#0	%0	M	SD	Med	IQR	R	#0	%0
Subjective intoxication immediately after beverage consumption
A/D	2.49*	0.79	3	1	0–4	1*	2	1.47*a	0.86	2	1	0–3	5*	14
D-Only	2.58*	0.81	3	1	1–4	0	0	1.48*a	0.79	1	1	0–4	2*	6
Collapsing WM Updating Task Completion condition	2.58*	0.80	3	1	0–4	1*	1	1.48*a	0.82	1	1	0–4	7*a	10
Subjective intoxication during AL procedures
A/D	2.29*	0.79	2	1	0–4	2*	4	1.35*a	0.77	1	1	0–3	5*	14
D-Only	2.79*	0.81	3	1	1–4	0	0	1.33*a	0.92	1	1	0–3	6*a	17
Collapsing WM Updating Task Completion condition	2.52*	0.83	2	1	0–4	2*	2	1.34*a	0.84	1	1	0–3	11*a	15
Subjective intoxication during DL procedures
A/D	1.55	0.78	1	1	0–4	2*	4	0.68*a	0.81	0.5	1	0–3	17a	47
D-Only	1.92	0.85	2	1	0–4	1*	3	0.88*a	0.78	1	1	0–3	11a	31
Collapsing WM Updating Task Completion condition	1.72*	0.83	2	1	0–4	3*	3	0.78*a	0.79	1	1	0–3	28a	39
Perceived number of standard drink equivalents consumed
A/D	4.09*	1.18	4	2	3–8	0	0	2.62*a	1.30	3	1	0–7	2*	5
D-Only	4.10*	1.20	4	2	2–7	0	0	2.81*a	1.50	3	1	1–7	0	0
Collapsing WM Updating Task Completion condition	4.09*	1.19	4	2	2–8	0	0	2.71*a	1.39	3	1	0–7	2*	3

Note. A/D = participants who completed the WM Updating tasks on both the ascending limb (AL) and descending limb (DL) of the breath alcohol concentration curve (Alcohol n = 46; Placebo n = 36). D-Only = participants who completed the WM Updating tasks only on the DL (Alcohol n = 39; Placebo n = 35). Med = median; IQR = inter-quartile range. R = range, shown as min – max. #0 = Number of participants for whom response was 0. %0 = percentage of participants for whom response was 0. Item and response language are provided in Supplemental Materials. Mean drink estimates and intoxication ratings were evaluated using t tests (* = p < .05 for null hypothesis test of group mean equivalence to 0; ^a = p < .05 for null hypothesis test of Placebo group mean equivalence to Alcohol group mean). Rank-based Wilcoxon tests (on group medians) produced the same pattern of results. Pearson χ² tests (df = 1) were used on the #0 (* = p < .05 for null hypothesis test of uniform distribution of #0s and #non-0s within groups; ^a = p < .05 for null hypothesis test of similar distribution of #0s and #non-0s between Placebo and Alcohol groups). For drink estimate #0, the null hypothesis test of uniformly distributed 0s and non-0s could not be conducted for experimental cells in which no participants indicated 0 drinks.

WM Updating Performance

Figure 3 presents the baseline,^iv AL, and DL mean accuracy for each WM Updating task as a function of beverage and task completion conditions.

Figure 3. Mean Accuracy Across Testing Occasions for Each Task as a Function of Beverage and Session 2 Task Completion Condition.

Note. Transformed accuracy is the arcsine of the square root of the winsorized proportion correct. Testing occasion Baseline refers to session 1. Testing occasion AL and DL are the ascending limb and descending limb of the BAC curve, respectively, or corresponding timepoints during session 2. Tasks were completed on DL only (D-Only) or on both AL and DL (A/D). Beverage conditions were: alcohol, placebo, or control. Sample sizes were 46, 36, 38, 39, 35, and 37 for Alcohol A/D, Placebo A/D, Control A/D, Alcohol D-Only, Placebo D-Only, and Control D-Only groups, respectively. Error bars are ± 1 SEM.

A model with equal task factor loadings and intercepts across groups and testing occasions, and equal residual task variances across groups fit the data well (χ²=83, df=59, CFI=.961, TLI=.953, RMSEA=.059; see Table S5 for invariance analyses and Table S6 for final model parameters).^v

Acute Effects of Alcohol

Alcohol v. No Alcohol Groups

Baseline latent means did not differ across groups Alcohol and No Alcohol, but there were group differences on AL and DL (Table 7, Figure 4; see Table S7 for equality tests). On both AL and DL, latent means were higher for group No Alcohol compared to Alcohol.

Table 7.

Latent WM Updating Means Across Groups and Testing Occasions

Alcohol Effects Model (N = 231)	No Alcohol (n = 146)			Alcohol (n = 85)
WM Updating	M	SE	p	M	SE	p
Baseline	.00	.00	---	.00	.01	.634
Ascending limb	.05	.01	<.001	.01	.01	.560
Descending limb	.08	.01	<.001	.05	.01	<.001
Expectancy Effects Model (N = 231)	No Expectancy (n = 75)			Expectancy (n = 156)
WM Updating	M	SE	p	M	SE	p
Baseline	.00	.00	---	.00	.01	.705
Ascending limb	.05	.01	<.001	.03	.01	.010
Descending limb	.08	.01	<.001	.07	.01	<.001
Pure Expectancy Effects Model (N = 146)	No Expectancy (n = 75)			Expectancy (n = 71)
WM Updating	M	SE	p	M	SE	p
Baseline	.00	.00	---	.00	.01	1.00
Ascending limb	.05	.01	<.001	.05	.01	<.001
Descending limb	.08	.01	<.001	.09	.02	<.001

Note. All means were estimated from models under strict measurement invariance constraints. P values indicate the significance of change from the reference level, which was Baseline WM Updating in group No Alcohol or No Expectancy, depending on the model.

Figure 4. Latent WM Updating Factor Means Across Testing Occasions for Alcohol v. No Alcohol and Expectancy v. No Expectancy Models.

Note. Testing occasion Baseline refers to session 1. Testing occasion AL and DL are the ascending limb and descending limb of the BAC curve, respectively, or corresponding timepoints during session 2. Figure legend shows experiment cells pooled into each group in the multigroup latent variable model. A/D and D in the figure legend refer to session 2 assessment on both AL and DL or DL only, respectively. Baseline WM Updating in the No Alcohol or No Expectancy group served as the reference factor. Error bars are ± 1 SE.

Expectancy v. No Expectancy Groups

When the alcohol and placebo conditions were pooled into group Expectancy, Baseline and DL latent means did not differ between groups Expectancy and No Expectancy, but there was a group difference on AL (Table 7, Figure 4; see Table S7 for equality tests). On AL, latent means were higher for group No Expectancy relative to Expectancy. However, when group Expectancy contained only the placebo condition, there were no differences in WM Updating between groups Expectancy and No Expectancy at Baseline, AL, or DL (Table 7, Figure 4; see Table S7 for equality tests). Together, these findings indicate that the expectancy effect in the traditional expectancy model that pooled alcohol and placebo conditions was a false positive reflecting the pharmacological effect of alcohol, and that there was no “pure” expectancy effect.

Differences Across Limbs of the BAC Curve

Alcohol v. No Alcohol Groups

For both groups, latent means increased across occasions, but the pattern of means differed within groups (Table 7, Figure 4; see Table S7 for equality tests). In group No Alcohol, latent means were significantly different among all testing occasions, such that: Baseline < AL < DL. In group Alcohol, the latent mean was not significantly different at Baseline compared to AL but was significantly lower at Baseline compared to DL, and at AL compared to DL.

Expectancy v. No Expectancy Groups

For both groups, latent means increased across occasions, and in both groups, all latent means differed from each other (Baseline < AL < DL; Table 7, Figure 4; see Table S7 for equality tests).

Acute Tolerance

Alcohol v. No Alcohol Groups

To determine whether improvement across post-consumption testing occasions in group Alcohol was due to acute tolerance, we tested whether Task Completion condition (0=A/D, 1=D-Only) predicted DL WM Updating differently across Alcohol and No Alcohol groups (Table S8). Task Completion condition was negatively associated with DL WM Updating to the same extent in both groups, indicating lower WM Updating performance when tasks were completed only once (on DL or corresponding timepoints). This suggests that acute tolerance does not account for improved performance of group Alcohol at DL relative to AL.

Next, we re-estimated the latent means at each occasion after dropping D-Only participants (Table S9) to examine whether Alcohol and No Alcohol groups differed in the degree of performance improvement across testing occasions. WM Updating latent means were significantly higher at AL than at Baseline in group No Alcohol, but Baseline and AL latent means were equivalent in group Alcohol. This suggests that alcohol prevented performance improvement from Baseline to AL. In both groups, latent means were lower at AL than at DL. Latent means at DL were lower in group Alcohol compared to No Alcohol, but to a lesser degree than when D-Only participants were included in both groups. Together, these findings indicate that alcohol also attenuated performance improvement from AL to DL.

Discussion

By administering multiple laboratory tasks across separate sessions to a large sample, including both placebo and no-alcohol control conditions, and comparing performance across BAC curve limbs, the current study provided the most comprehensive test to date of alcohol’s acute effects on WM Updating. We found both between- and within-subject effects of alcohol pharmacology on WM Updating. Across time, WM Updating was lower among intoxicated compared with sober persons, consistent with the previously reported between-subject effect on WM tasks.^13,66,67 Nonetheless, alcohol consumption did not diminish participants’ WM Updating performance relative to baseline ability. This finding is consistent with older reports^68,69 but is inconsistent with more recent reports.^17,22,70 Moreover, we found no between- or within-subject effects of alcohol expectancy on WM Updating. ^vi One possible explanation is that participants held no strong expectancies about negative effects of alcohol on task performance, and so did not attempt to compensate for anticipated impairment.²⁸ Another possibility is that although the placebo manipulation convinced nearly all participants that they had consumed alcohol, they did not feel intoxicated enough to warrant compensatory efforts.

We also found that WM Updating performance was better on the DL compared with baseline or the AL, but there was essentially no evidence that DL relative to AL performance in the alcohol group was due to acute tolerance rather than repeated testing effects.^vii An acute tolerance account^32,33 would predict poorer performance on the AL among intoxicated relative to sober individuals, but sober-equivalent—or, at least, significantly improved—performance among intoxicated individuals on the DL. Instead, intoxicated individuals performed more poorly at both timepoints, relative to their sober counterparts. Furthermore, individuals tested only on the DL exhibited poorer performance on the DL than counterparts in the AL/DL condition, but that deficit did not differ between alcohol and no-alcohol groups. Thus, repeated testing effects were present in sober and intoxicated states, albeit attenuated somewhat in the latter. Taken together, our findings indicate that at a dose level sufficient to produce a peak BAC of .085 g% within 30 min, alcohol’s acute effect on WM Updating appears to manifest as attenuation of improvements in performance otherwise experienced upon repeated testing.^viii

The current findings should be considered in light of the study’s limitations. First, although a relatively large dose was administered, within-person impairment and acute tolerance effects were not observed. One possible explanation is that participants’ typical peak BAC impacted the opportunity to observe alcohol-induced impairment, and thus, acute tolerance. Arguing against this possibility, the latent mean pattern and the test for acute tolerance were unchanged by adjustment for between-person differences in typical alcohol use (a proxy for between-person differences in typical peak BAC; see Tables S19–20). Nonetheless, acute impairment and acute tolerance still might be observed at higher doses characterizing typical drinking experiences for high-intensity drinkers (e.g., ≥ 0.10 g/kg).^22,67 Second, the tasks used here relied on the same WM Updating subprocesses (i.e., retrieval and substitution, but not transformation),⁵⁸ and the study was not designed for subprocess dissociation, so it remains to be seen whether different subprocesses are similarly (un)affected by alcohol. Third, WM Updating tasks were scored for accuracy but not response time (RT)—largely because they are not structured as RT tasks. Acute tolerance effects are observed more readily on response activation (indexed by RT) than mnemonic (indexed by accuracy) processes.^34,35,71 However, alcohol-induced cognitive or psychomotor slowing is not unique to WM Updating tasks, and its presence would not change the interpretation of the current findings. Fourth, acute alcohol can affect cognitive abilities differently depending upon between-person differences in baseline (i.e., sober state) ability.^27,52,53,72 In some cases, higher baselines may “buffer” against acute insult.⁷³ Alternatively, individuals with higher sober state ability may have “more to lose” from acute insult and could regress to population-mean levels of EF when intoxicated.^74,75 Regression analyses (see the Supplemental Materials) found no evidence that baseline ability moderated the acute effect of alcohol at the level of each task (Tables S3, S12). Nonetheless, latent variable models examining moderation were not viable, so it remains unclear whether sober state WM Updating ability could moderate the acute effect of alcohol on the common variance across WM Updating tasks. Finally, regression analyses, repeated-measures ANOVAs, and within-subject AL – baseline and DL – baseline difference score analyses (see Supplemental Materials) converged with the latent WM Updating analyses, but also suggested that performance on the verbal WM tasks (Keep Track and Letter Memory) adhered more closely to the pattern of effects detected in the latent WM Updating analyses than did performance on the visuospatial WM task (Spatial 2-Back). Moreover, those supplemental analyses suggested that alcohol attenuated performance improvements more strongly in the verbal WM tasks compared to the visuospatial WM task. Differential sensitivity of verbal vs. visuospatial WM to acute effects of alcohol is consistent with findings from a recent literature review,¹² but remains to be verified in a future study that can model WM subtypes (e.g., one using multiple verbal and visuospatial tasks).

Our findings have broader implications for understanding the acute effects of alcohol on EF and their implications for drinking-related negative consequences. Contemporary models emphasize “unity and diversity” in EF,⁷⁶ with unity captured by a Common factor onto which tasks from all facets load, and diversity captured by residual Shifting- and WM Updating-specific factors.^5,6,77 It is possible that alcohol’s impairment of different facets of EF reflects acute effects on the Common factor.^ix This possibility is bolstered by congruence between the current findings and findings of our previous work in independent samples testing acute alcohol effects on Inhibition⁵² and Shifting.⁵³ Testing this idea in future studies requires modeling alcohol’s acute effects on multiple EF facets, as well as on Common EF, in the same individuals.

Although alcohol prevented the practice effects observed in the control and placebo conditions, it did not produce an absolute decrease in WM Updating performance in our study. Such a decrease might be found in future studies employing experimental designs that, unlike ours, do not aim to evaluate potential practice effects but rather to overcome them (e.g., by having participants practice the tasks until asymptotic performance before beverage administration). Alcohol-induced deficits in WM Updating performance could encourage excessive drinking, either through failure to keep track of drinks consumed as a drinking episode unfolds or through failure to maintain personal drinking reduction goals. Either process could contribute to the experience of negative consequences, including the potential to prolong or worsen AUD. These deficits also may limit the extent or duration of an individual’s benefit from interventions like motivational interviewing⁷⁸ that rely on individuals maintaining awareness of set goals (e.g., drinking reduction) and implementing behavioral strategies to attain them (e.g., decreasing drinking episode frequency and the number of drinks consumed within episodes).

In conclusion, the acute effects of alcohol on WM Updating attenuated performance improvement otherwise experienced upon repeated testing. Future studies should examine not only the acute effect of alcohol on other EF facets (e.g., access to, or strategic retrieval of, information in long-term memory), but also the extent to which similarity in the acute effect of alcohol on different EF facets can be accounted for by an acute effect of alcohol on the theorized Common factor in EF.^5,6,77 Characterizing the acute effects of alcohol on EF is an important step toward understanding the cognitive-behavioral mechanisms that underlie problematic alcohol use.