I representation: Compiled Results

Steven

2025-07-31

Q1) How does “I” representation change across time

Good vs Bad

I doesn’t move from bad to good and the differences are very small.

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: i_MADS ~ days_c + days_c2 + (1 | room_id)
   Data: good_bad_data_small

REML criterion at convergence: 26984.8

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-3.8546 -0.5424 -0.0288  0.5117  6.4138 

Random effects:
 Groups   Name        Variance Std.Dev.
 room_id  (Intercept) 0.5301   0.7281  
 Residual             0.4742   0.6886  
Number of obs: 10450, groups:  room_id, 3627

Fixed effects:
              Estimate Std. Error         df t value Pr(>|t|)    
(Intercept) -7.431e-04  1.571e-02  5.405e+03  -0.047    0.962    
days_c       4.067e-02  7.284e-03  7.547e+03   5.584 2.43e-08 ***
days_c2     -4.067e-04  7.322e-03  7.414e+03  -0.056    0.956    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
        (Intr) days_c
days_c   0.067       
days_c2 -0.453 -0.129

Able vs Unable

I moves from unable to able

Q2) How does “I” representation along these axes relate to symptoms?

Good vs Bad

More “good” representation = less sx (both between- and within-person)

Able vs Unable

More “able” representation = less sx (both between- and within-person)

Q3) Are these results attributable to a set of potential confounding variables?

Good vs Bad

Word Frequency

Controlling for all the Words Frequency and Total words

No, all results hold when controlling for frequency of “I” and other words

Sentiment, Emotion Words, and Total Word Count

No, all results hold when controlling for sentiment and emotion words and total WC.

Able vs Unable

Controlling for all the Words Frequency and Total words

Sentiment, Emotion Words, and Total Word Count

Q4) Is this occurring for all words, or is “I” unique in some way?

No, control words ** the, and, to, be, a, we ** do not uniformly move along these axes and/or they do not relate to symptoms.

Frequencies

Changes over Therapy

Track with Changes in Symptoms

Comparing to ‘I’ Models

Good

Able

Q5) Descriptive Approaches

Story of I DLATK Analyses

1st Half of Treatment

Used median split for daysSinceFirstText to create two groups Median treatment time = 35 days.

2nd Half of Treatment

Topic Modeling

Built topic model from all sentences in Talksapace data using bertopic from sentences that had the word I in them.
Extracted the top 20 topics from the model

Supplemental Analyses

Examining other pairs

Looking at happy-sad, calm-anxious, can-cannot.

Changes over Therapy

Track with Changes in Symptoms

All models control for time in tx and are specified separately for each word.

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
Internalizing ~ daysSinceFirstText + can_MADS_within + can_MADS_between +  
    pct_can_within + pct_can_between + +emo_pos_within + emo_pos_between +  
    emo_neg_within + emo_neg_between + pct_i_within + pct_i_between +  
    tone_pos_within + tone_pos_between + tone_neg_within + tone_neg_between +  
    n_total_within + n_total_between + (1 | room_id)
   Data: disaggregated

REML criterion at convergence: 26537.6

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-3.4321 -0.5639 -0.0939  0.4863  4.7030 

Random effects:
 Groups   Name        Variance Std.Dev.
 room_id  (Intercept) 0.5107   0.7146  
 Residual             0.3589   0.5991  
Number of obs: 11315, groups:  room_id, 3409

Fixed effects:
                     Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)        -1.544e-03  1.351e-02  3.394e+03  -0.114 0.908999    
daysSinceFirstText -2.412e-01  6.379e-03  8.419e+03 -37.811  < 2e-16 ***
can_MADS_within    -4.961e-02  7.114e-03  7.949e+03  -6.973 3.35e-12 ***
can_MADS_between    4.816e-02  1.406e-02  3.422e+03   3.424 0.000623 ***
pct_can_within      3.315e-03  6.751e-03  7.899e+03   0.491 0.623399    
pct_can_between    -5.857e-03  1.371e-02  3.458e+03  -0.427 0.669147    
emo_pos_within     -2.543e-02  8.340e-03  7.899e+03  -3.049 0.002306 ** 
emo_pos_between    -8.367e-02  1.929e-02  3.478e+03  -4.337 1.48e-05 ***
emo_neg_within      3.397e-02  1.069e-02  7.899e+03   3.178 0.001487 ** 
emo_neg_between    -2.377e-02  3.271e-02  3.421e+03  -0.727 0.467565    
pct_i_within        4.422e-03  6.810e-03  7.899e+03   0.649 0.516111    
pct_i_between       3.475e-02  1.392e-02  3.444e+03   2.496 0.012602 *  
tone_pos_within    -3.480e-02  8.430e-03  7.900e+03  -4.128 3.70e-05 ***
tone_pos_between   -2.797e-02  1.953e-02  3.494e+03  -1.432 0.152148    
tone_neg_within     1.198e-02  1.086e-02  7.903e+03   1.103 0.270108    
tone_neg_between    2.203e-01  3.282e-02  3.421e+03   6.714 2.21e-11 ***
n_total_within     -1.178e-02  6.809e-03  7.899e+03  -1.730 0.083715 .  
n_total_between     1.833e-02  1.375e-02  3.417e+03   1.333 0.182747    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1