Updating predictions by previous prediction error
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ pe_2_lag1 + exam + (1 | cohort/id)
## Data: df.new
##
## REML criterion at convergence: 12583.5
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.8291 -0.5280 -0.0194 0.5468 3.8890
##
## Random effects:
## Groups Name Variance Std.Dev.
## id:cohort (Intercept) 0.000 0.00
## cohort (Intercept) 3.881 1.97
## Residual 222.281 14.91
## Number of obs: 1526, groups: id:cohort, 696; cohort, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 1.8212 1.7454 8.8016 1.043 0.325
## pe_2_lag1 0.3027 0.0317 1522.9993 9.550 <2e-16 ***
## exam -0.3934 0.4144 1469.2917 -0.949 0.343
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) p_2_l1
## pe_2_lag1 0.003
## exam -0.722 -0.046
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
## Warning: Removed 2839 rows containing missing values (geom_point).
Splitting by PE direction
Positive PEs drive prediction updating
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ pe_2_lag1 + (1 | id)
## Data: df.new[which(df.new$pe_2_lag1 > 0), ]
##
## REML criterion at convergence: 7357.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.7812 -0.5184 -0.0199 0.5320 3.6231
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0.00
## Residual 232.1 15.23
## Number of obs: 888, groups: id, 521
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -2.36073 0.89791 886.00000 -2.629 0.00871 **
## pe_2_lag1 0.51091 0.07134 886.00000 7.161 1.68e-12 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## pe_2_lag1 -0.822
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
## Warning: Removed 503 rows containing missing values (geom_point).
Takeaway: model positive and negative PEs separately
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ unsigned_lag_pe_2 * pe_2_lag_sign_f + (1 | id)
## Data: df.new
##
## REML criterion at convergence: 12400.7
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.8526 -0.5162 0.0646 0.5327 3.8085
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0.00
## Residual 225.3 15.01
## Number of obs: 1502, groups: id, 694
##
## Fixed effects:
## Estimate Std. Error df t value
## (Intercept) -2.12677 1.00184 1498.00000 -2.123
## unsigned_lag_pe_2 -0.03783 0.08232 1498.00000 -0.460
## pe_2_lag_sign_f1 -0.23396 1.33655 1498.00000 -0.175
## unsigned_lag_pe_2:pe_2_lag_sign_f1 0.54874 0.10825 1498.00000 5.069
## Pr(>|t|)
## (Intercept) 0.0339 *
## unsigned_lag_pe_2 0.6459
## pe_2_lag_sign_f1 0.8611
## unsigned_lag_pe_2:pe_2_lag_sign_f1 4.49e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) un___2 p_2___
## unsgnd_l__2 -0.797
## p_2_lg_sg_1 -0.750 0.597
## u___2:_2___ 0.606 -0.760 -0.807
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
Computing information content
## Warning: Removed 512 row(s) containing missing values (geom_path).
Updating vs. info content
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ log_info_content_lag1 * pe_2_lag_sign_f + (1 | id)
## Data: df.new
##
## REML criterion at convergence: 4662.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.4666 -0.5300 -0.0745 0.5864 3.7737
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0.00
## Residual 179.8 13.41
## Number of obs: 582, groups: id, 248
##
## Fixed effects:
## Estimate Std. Error df t value
## (Intercept) -13.210 2.446 578.000 -5.401
## log_info_content_lag1 5.517 1.133 578.000 4.870
## pe_2_lag_sign_f1 14.661 3.265 578.000 4.490
## log_info_content_lag1:pe_2_lag_sign_f1 -5.131 1.540 578.000 -3.332
## Pr(>|t|)
## (Intercept) 9.70e-08 ***
## log_info_content_lag1 1.44e-06 ***
## pe_2_lag_sign_f1 8.60e-06 ***
## log_info_content_lag1:pe_2_lag_sign_f1 0.000917 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) lg___1 p_2___
## lg_nf_cnt_1 -0.934
## p_2_lg_sg_1 -0.749 0.700
## l___1:_2___ 0.687 -0.736 -0.938
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
possible explanation: PEs cause updating in expected direction when information content is low (i.e., surprise is minimal)
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula:
## delta_pred_2 ~ log_info_content_lag1 * pe_2_lag_sign_f + unsigned_lag_pe_2 *
## pe_2_lag_sign_f + (1 | id)
## Data: df.new
##
## REML criterion at convergence: 4629
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.6030 -0.5287 0.0538 0.5521 3.5874
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0
## Residual 168.9 13
## Number of obs: 582, groups: id, 248
##
## Fixed effects:
## Estimate Std. Error df t value
## (Intercept) -13.1000 2.3757 576.0000 -5.514
## log_info_content_lag1 5.8892 1.2086 576.0000 4.873
## pe_2_lag_sign_f1 10.4074 3.2382 576.0000 3.214
## unsigned_lag_pe_2 -0.1019 0.1384 576.0000 -0.737
## log_info_content_lag1:pe_2_lag_sign_f1 -6.8260 1.5899 576.0000 -4.293
## pe_2_lag_sign_f1:unsigned_lag_pe_2 0.8399 0.1824 576.0000 4.604
## Pr(>|t|)
## (Intercept) 5.30e-08 ***
## log_info_content_lag1 1.42e-06 ***
## pe_2_lag_sign_f1 0.00138 **
## unsigned_lag_pe_2 0.46162
## log_info_content_lag1:pe_2_lag_sign_f1 2.07e-05 ***
## pe_2_lag_sign_f1:unsigned_lag_pe_2 5.10e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) lg___1 p_2_l__1 un___2 l___1:
## lg_nf_cnt_1 -0.821
## p_2_lg_sg_1 -0.734 0.602
## unsgnd_l__2 -0.063 -0.418 0.046
## l___1:_2___ 0.624 -0.760 -0.818 0.318
## p_2___1:___ 0.048 0.317 -0.169 -0.759 -0.328
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
see if information content or PE differentially predict updating for pos vs. neg PEs
Need to try fitting curves to these data…
Positive PEs
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ pe_2_lag1 * log_info_content_lag1 + (1 | id)
## Data: df.new[which(df.new$pe_2_lag1 > 0), ]
##
## REML criterion at convergence: 2728.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0375 -0.5560 0.0138 0.5408 3.7640
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0.00
## Residual 153.3 12.38
## Number of obs: 347, groups: id, 202
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -2.20901 4.48982 343.00000 -0.492 0.6230
## pe_2_lag1 0.69646 0.35895 343.00000 1.940 0.0532
## log_info_content_lag1 -1.19568 2.34226 343.00000 -0.510 0.6100
## pe_2_lag1:log_info_content_lag1 0.02146 0.17619 343.00000 0.122 0.9031
##
## (Intercept)
## pe_2_lag1 .
## log_info_content_lag1
## pe_2_lag1:log_info_content_lag1
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) p_2_l1 lg___1
## pe_2_lag1 -0.884
## lg_nf_cnt_1 -0.963 0.834
## p_2_l1:___1 0.884 -0.949 -0.907
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
Negative PEs
## boundary (singular) fit: see ?isSingular
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: delta_pred_2 ~ pe_2_lag1 * log_info_content_lag1 + (1 | id)
## Data: df.new[which(df.new$pe_2_lag1 < 0), ]
##
## REML criterion at convergence: 1898.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3601 -0.4846 0.1823 0.5959 2.3758
##
## Random effects:
## Groups Name Variance Std.Dev.
## id (Intercept) 0.0 0.00
## Residual 192.2 13.86
## Number of obs: 235, groups: id, 152
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -18.3568 4.8360 231.0000 -3.796 0.000188
## pe_2_lag1 -0.3244 0.3652 231.0000 -0.888 0.375334
## log_info_content_lag1 8.5357 2.4418 231.0000 3.496 0.000567
## pe_2_lag1:log_info_content_lag1 0.1916 0.1501 231.0000 1.276 0.203145
##
## (Intercept) ***
## pe_2_lag1
## log_info_content_lag1 ***
## pe_2_lag1:log_info_content_lag1
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) p_2_l1 lg___1
## pe_2_lag1 0.792
## lg_nf_cnt_1 -0.950 -0.688
## p_2_l1:___1 -0.852 -0.915 0.849
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see ?isSingular
