Chapter 27 [fMRI] VPS ~ singletrial

What is the purpose of this notebook?

Here, I model vPS dot products as a function of cue, stimulus intensity and expectation ratings.
One of the findings is that low cues lead to higher VPS dotproducts in the high intensity group, and that this effect becomes non-significant across sessions.
03/23/2023: For now, I’m grabbing participants that have complete data, i.e. 18 runs, all three sessions. N = 65

## [1] "/Users/h/Dropbox (Dartmouth College)/projects_dropbox/social_influence_analysis/analysis/fmri/nilearn/signature_extract/signature-VPS_sub-all_runtype-pvc_event-stimulus.tsv"

27.1 VPS ~ 3 task * 3 stimulus_intensity

Q. What does the VPS pattern look like for the three tasks?

model.alltask <-
  lmer(VPS ~ task_con_linear*stim_factor + 
         task_con_quad*stim_factor + (task|sub), data = pvc)
sjPlot::tab_model(model.alltask)

	VPS
Predictors	Estimates	CI	p
(Intercept)	1.24	0.65 – 1.83	<0.001
task con linear	-5.30	-7.15 – -3.46	<0.001
stim factor [low]	0.29	-0.01 – 0.59	0.056
stim factor [med]	0.23	-0.07 – 0.53	0.136
task con quad	-2.11	-3.90 – -0.32	0.021
task con linear * stim factor [low]	-0.59	-2.02 – 0.84	0.421
task con linear * stim factor [med]	-1.54	-2.98 – -0.11	0.035
stim factor [low] * task con quad	2.61	1.36 – 3.87	<0.001
stim factor [med] * task con quad	2.21	0.96 – 3.46	0.001
Random Effects
σ²	74.67
τ₀₀ _sub	16.59
τ₁₁ _sub.taskpain	35.89
τ₁₁ _{sub.taskvicarious}	16.74
ρ₀₁	-0.51
	-0.68
ICC	0.18
N _sub	111
Observations	19428
Marginal R² / Conditional R²	0.072 / 0.241

eta-squared

Parameter	Eta2_partial	CI	CI_low	CI_high
task_con_linear	0.3266266	0.95	0.2104200	1
stim_factor	0.0002117	0.95	0.0000000	1
task_con_quad	0.0034213	0.95	0.0000000	1
task_con_linear:stim_factor	0.0002379	0.95	0.0000000	1
stim_factor:task_con_quad	0.0010135	0.95	0.0003561	1

Cohen’s d

	t	df	d
task_con_linear	-5.6291153	164.6488	-0.8773863
stim_factorlow	1.9136684	19090.4418	0.0277006
stim_factormed	1.4905241	19090.4418	0.0215755
task_con_quad	-2.3069934	148.6413	-0.3784483
task_con_linear:stim_factorlow	-0.8050546	19090.4422	-0.0116533
task_con_linear:stim_factormed	-2.1116090	19090.4422	-0.0305658
stim_factorlow:task_con_quad	4.0883679	19090.4422	0.0591796
stim_factormed:task_con_quad	3.4549197	19090.4422	0.0500104

27.2 VPS ~ vicarious task: 2 cue x 3 stimulus_intensity

Q. Within vicarious task, Does stimulus intenisty level and cue level significantly predict VPS dotproducts?

27.2.1 Linear model ver 1: VPS ~ stimulus_intensity

model.vpsstim <- lmer(VPS ~ stim_con_linear + stim_con_quad +(stim_con_linear + stim_con_quad |sub), data = data_screen)
sjPlot::tab_model(model.vpsstim)

	VPS
Predictors	Estimates	CI	p
(Intercept)	1.54	0.83 – 2.25	<0.001
stim con linear	0.39	-0.18 – 0.95	0.180
stim con quad	-0.23	-0.72 – 0.27	0.370
Random Effects
σ²	55.74
τ₀₀ _sub	12.05
τ₁₁ _{sub.stim_con_linear}	0.65
τ₁₁ _{sub.stim_con_quad}	0.44
ρ₀₁	0.56
	-0.18
N _sub	104
Observations	4420
Marginal R² / Conditional R²	0.001 / NA

27.2.1.1 Linear model ver 1: eta-squared

Parameter	Eta2_partial	CI	CI_low	CI_high
stim_con_linear	0.0112309	0.95	0	1
stim_con_quad	0.0054742	0.95	0	1

Linear model ver 1: Cohen’s d VPS stimulus intensity = 0.213

	t	df	d
stim_con_linear	1.3395560	157.9802	0.2131521
stim_con_quad	-0.8965857	146.0413	-0.1483829

27.2.2 Linear model ver 2: VPS ~ stimulus_intenisy * cue

model.vpscuestim <- lmer(VPS ~ cue_con*stim_con_linear + cue_con*stim_con_quad + (cue_con+stim_factor|sub), data = data_screen)
sjPlot::tab_model(model.vpscuestim)

	VPS
Predictors	Estimates	CI	p
(Intercept)	1.54	0.83 – 2.25	<0.001
cue con	0.08	-0.36 – 0.53	0.709
stim con linear	0.39	-0.18 – 0.95	0.179
stim con quad	-0.23	-0.72 – 0.27	0.369
cue con * stim con linear	0.03	-1.05 – 1.10	0.961
cue con * stim con quad	-0.09	-1.03 – 0.86	0.858
Random Effects
σ²	55.78
τ₀₀ _sub	14.25
τ₁₁ _{sub.cue_con}	0.02
τ₁₁ _{sub.stim_factorlow}	0.66
τ₁₁ _{sub.stim_factormed}	1.08
ρ₀₁	0.97
	-0.67
	-0.47
N _sub	104
Observations	4420
Marginal R² / Conditional R²	0.001 / NA

Linear model ver 2: eta-squared

Parameter	Eta2_partial	CI	CI_high
cue_con	0.0000478	0.95	1
stim_con_linear	0.0116418	0.95	1
stim_con_quad	0.0057814	0.95	1
cue_con:stim_con_linear	0.0000006	0.95	1
cue_con:stim_con_quad	0.0000076	0.95	1

Cohen’s d: VPS stimulus intensity d = 0.217, cue level d = 0.013

	t	df	d
cue_con	0.3735172	2920.3096	0.0138238
stim_con_linear	1.3445086	153.4694	0.2170614
stim_con_quad	-0.8976592	138.5715	-0.1525122
cue_con:stim_con_linear	0.0484256	4231.8238	0.0014888
cue_con:stim_con_quad	-0.1789583	4233.5751	-0.0055008

Lineplots

27.2.3 2 Cue x 3 Stimulus intensity x Expectation rating

model.vps3factor <- lmer(VPS ~ cue_con*stim_con_linear*event02_expect_angle + cue_con*stim_con_quad*event02_expect_angle + (cue_con|sub), data = data_screen)
sjPlot::tab_model(model.vps3factor)

	VPS
Predictors	Estimates	CI	p
(Intercept)	1.24	0.40 – 2.08	0.004
cue con	-0.15	-1.09 – 0.80	0.758
stim con linear	0.32	-0.77 – 1.41	0.564
event02 expect angle	0.01	-0.00 – 0.03	0.158
stim con quad	-0.14	-1.09 – 0.82	0.781
cue con * stim con linear	0.33	-1.85 – 2.51	0.768
cue con * event02 expect angle	-0.00	-0.03 – 0.03	0.839
stim con linear * event02 expect angle	0.01	-0.02 – 0.04	0.624
cue con * stim con quad	0.33	-1.58 – 2.24	0.738
event02 expect angle * stim con quad	-0.01	-0.03 – 0.02	0.707
(cue con * stim con linear) * event02 expect angle	-0.02	-0.09 – 0.05	0.544
(cue con * event02 expect angle) * stim con quad	0.00	-0.05 – 0.06	0.990
Random Effects
σ²	56.38
τ₀₀ _sub	12.03
τ₁₁ _{sub.cue_con}	0.02
ρ₀₁ _sub	1.00
N _sub	104
Observations	4230
Marginal R² / Conditional R²	0.002 / NA

kableExtra::kable_styling(
  knitr::kable(
    eta_squared(model.vps3factor, partial = TRUE), # MODIFY
    "html"), "striped", position = "left", font_size = 12)

Parameter	Eta2_partial	CI	CI_high
cue_con	0.0000281	0.95	1
stim_con_linear	0.0000805	0.95	1
event02_expect_angle	0.0005784	0.95	1
stim_con_quad	0.0000187	0.95	1
cue_con:stim_con_linear	0.0000210	0.95	1
cue_con:event02_expect_angle	0.0000121	0.95	1
stim_con_linear:event02_expect_angle	0.0000582	0.95	1
cue_con:stim_con_quad	0.0000271	0.95	1
event02_expect_angle:stim_con_quad	0.0000342	0.95	1
cue_con:stim_con_linear:event02_expect_angle	0.0000892	0.95	1
cue_con:event02_expect_angle:stim_con_quad	0.0000000	0.95	1

kableExtra::kable_styling(
  knitr::kable(
    lme.dscore(model.vps3factor, data_screen, type = "lme4"), # MODIFY
    "html"), "striped", position = "left", font_size = 12)

	t	df	d
cue_con	-0.3079865	3375.733	-0.0106017
stim_con_linear	0.5762780	4122.869	0.0179499
event02_expect_angle	1.4107477	3438.911	0.0481137
stim_con_quad	-0.2776990	4120.037	-0.0086527
cue_con:stim_con_linear	0.2945420	4124.835	0.0091722
cue_con:event02_expect_angle	-0.2038101	3432.844	-0.0069571
stim_con_linear:event02_expect_angle	0.4897779	4124.654	0.0152523
cue_con:stim_con_quad	0.3339531	4120.459	0.0104050
event02_expect_angle:stim_con_quad	-0.3755729	4121.327	-0.0117005
cue_con:stim_con_linear:event02_expect_angle	-0.6068106	4126.822	-0.0188919
cue_con:event02_expect_angle:stim_con_quad	0.0120689	4121.074	0.0003760

27.3 VPS ~ session * cue * stimulus_intensity

Q. Is the cue effect on VPS different across sessions?

Quick answer: Yes, the cue effect in session 1 (for high intensity group) is significantly different; the dirrection flips in session 3; whereas this different becomes non significant in session 4.

Session 1: 2 cue * 3 stimulus_intensity

Session 3: 2 cue * 3 stimulus_intensity

Session 4: 2 cue * 3 stimulus_intensity

27.4 VPS ~ outcome_rating

Q. Do higher VPS values indicate higher outcome ratings? (Vicarious task only)

Yes, Higher VPS values are associated with higher outcome ratings. The pattern is slightly different from the NPS plots, when separated out into three panels of stimulus intensity levels. Neeed to ponder on these results

27.4.1 outcome_ratings * cue

27.4.2 outcome_ratings * stimulus_intensity * cue

27.5 VPS ~ expectation_rating

Q. What is the relationship betweeen expectation ratings & VPS? (Vicarious task only)

Do we see a linear effect between expectation rating and VPS dot products? Also, does this effect differ as a function of cue and stimulus intensity ratings, as is the case for behavioral ratings?

Quick answer: Kind of, expectation ratings predict VPS dotproducts; almost only in the low cue conditions across all stimulus intensity levels.