This is our guiding research question.
How do in-the-moment experiences of youth (i.e., challenge and relevance) and other momentary factors cultivate situational interest and engagement in STEM activities?
Overall engagement now includes hard working, concentrating, enjoy, and interest (this is also the case for the longer-term outcomes study).
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.76 | 0.23 | <.001 | |
challenge | 0.04 | 0.02 | .022 | |
relevance | 0.60 | 0.02 | <.001 | |
gender (M) | -0.05 | 0.06 | .391 | |
classroom_versus_field_enrichment | 0.07 | 0.05 | .181 | |
CLASS_comp | -0.00 | 0.02 | .910 | |
youth_activity_rc (Basic Skills Activity) | -0.12 | 0.06 | .053 | |
youth_activity_rc (Creating Product) | -0.10 | 0.07 | .136 | |
youth_activity_rc (Field Trip Speaker) | -0.01 | 0.11 | .949 | |
youth_activity_rc (Lab Activity) | -0.02 | 0.11 | .838 | |
youth_activity_rc (Program Staff Led) | -0.10 | 0.07 | .159 | |
overall_pre_interest | 0.07 | 0.04 | .047 | |
youth_development_overall | 0.01 | 0.01 | .095 | |
prop_attend | 0.23 | 0.20 | .262 | |
Random Parts | ||||
Nbeep_ID_new | 227 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.048 | |||
ICCparticipant_ID | 0.132 | |||
ICCprogram_ID | 0.021 | |||
Observations | 2428 | |||
R2 / Ω02 | .545 / .542 |
This model suggests that:
What if we remove youth activities?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.77 | 0.22 | <.001 | |
challenge | 0.04 | 0.02 | .025 | |
relevance | 0.60 | 0.02 | <.001 | |
gender (M) | -0.05 | 0.06 | .432 | |
classroom_versus_field_enrichment | 0.06 | 0.05 | .235 | |
CLASS_comp | -0.02 | 0.02 | .302 | |
overall_pre_interest | 0.07 | 0.04 | .043 | |
youth_development_overall | 0.02 | 0.01 | .074 | |
prop_attend | 0.23 | 0.20 | .249 | |
Random Parts | ||||
Nbeep_ID_new | 230 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.048 | |||
ICCparticipant_ID | 0.131 | |||
ICCprogram_ID | 0.025 | |||
Observations | 2457 | |||
R2 / Ω02 | .545 / .541 |
This model shows basically a similar story.
What if we remove challenge and relevance?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 1.96 | 0.36 | <.001 | |
gender (M) | 0.10 | 0.10 | .347 | |
classroom_versus_field_enrichment | 0.05 | 0.06 | .438 | |
CLASS_comp | 0.02 | 0.03 | .518 | |
youth_activity_rc (Basic Skills Activity) | -0.05 | 0.07 | .422 | |
youth_activity_rc (Creating Product) | 0.02 | 0.08 | .841 | |
youth_activity_rc (Field Trip Speaker) | 0.12 | 0.13 | .354 | |
youth_activity_rc (Lab Activity) | 0.00 | 0.12 | .983 | |
youth_activity_rc (Program Staff Led) | -0.03 | 0.08 | .660 | |
overall_pre_interest | 0.14 | 0.06 | .020 | |
youth_development_overall | 0.01 | 0.01 | .219 | |
prop_attend | 0.31 | 0.34 | .368 | |
Random Parts | ||||
Nbeep_ID_new | 227 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.035 | |||
ICCparticipant_ID | 0.325 | |||
ICCprogram_ID | 0.027 | |||
Observations | 2428 | |||
R2 / Ω02 | .453 / .444 |
These activities seem to make more sense now. What if we remove the random beep effect?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.75 | 0.22 | <.001 | |
gender (M) | -0.05 | 0.06 | .398 | |
challenge | 0.04 | 0.02 | .013 | |
relevance | 0.60 | 0.02 | <.001 | |
classroom_versus_field_enrichment | 0.09 | 0.04 | .036 | |
CLASS_comp | 0.00 | 0.02 | .827 | |
youth_activity_rc (Basic Skills Activity) | -0.12 | 0.05 | .017 | |
youth_activity_rc (Creating Product) | -0.13 | 0.06 | .023 | |
youth_activity_rc (Field Trip Speaker) | -0.00 | 0.08 | .954 | |
youth_activity_rc (Lab Activity) | -0.03 | 0.09 | .750 | |
youth_activity_rc (Program Staff Led) | -0.09 | 0.05 | .091 | |
overall_pre_interest | 0.07 | 0.04 | .054 | |
youth_development_overall | 0.01 | 0.01 | .096 | |
prop_attend | 0.22 | 0.20 | .271 | |
Random Parts | ||||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCparticipant_ID | 0.129 | |||
ICCprogram_ID | 0.027 | |||
Observations | 2428 | |||
R2 / Ω02 | .499 / .498 |
Not a huge difference. What if we remove some of the covariates / controls but otherwise run model 1?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.99 | 0.14 | <.001 | |
gender (M) | -0.05 | 0.06 | .415 | |
challenge | 0.06 | 0.02 | .002 | |
relevance | 0.59 | 0.02 | <.001 | |
classroom_versus_field_enrichment | 0.07 | 0.05 | .201 | |
youth_activity_rc (Basic Skills Activity) | -0.11 | 0.06 | .049 | |
youth_activity_rc (Creating Product) | -0.05 | 0.06 | .430 | |
youth_activity_rc (Field Trip Speaker) | 0.01 | 0.11 | .960 | |
youth_activity_rc (Lab Activity) | 0.00 | 0.11 | .982 | |
youth_activity_rc (Program Staff Led) | -0.07 | 0.06 | .316 | |
overall_pre_interest | 0.08 | 0.04 | .028 | |
Random Parts | ||||
Nbeep_ID_new | 235 | |||
Nparticipant_ID | 180 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.050 | |||
ICCparticipant_ID | 0.128 | |||
ICCprogram_ID | 0.019 | |||
Observations | 2583 | |||
R2 / Ω02 | .541 / .537 |
Basic skills have a negative relationship with interest; otherwise, not large changes. What if we remove challenge and relevance?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 2.32 | 0.21 | <.001 | |
gender (M) | 0.11 | 0.10 | .281 | |
classroom_versus_field_enrichment | 0.04 | 0.06 | .506 | |
youth_activity_rc (Basic Skills Activity) | -0.03 | 0.06 | .630 | |
youth_activity_rc (Creating Product) | 0.09 | 0.07 | .152 | |
youth_activity_rc (Field Trip Speaker) | 0.13 | 0.13 | .294 | |
youth_activity_rc (Lab Activity) | 0.05 | 0.12 | .672 | |
youth_activity_rc (Program Staff Led) | 0.01 | 0.07 | .890 | |
overall_pre_interest | 0.15 | 0.06 | .012 | |
Random Parts | ||||
Nbeep_ID_new | 235 | |||
Nparticipant_ID | 180 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.040 | |||
ICCparticipant_ID | 0.324 | |||
ICCprogram_ID | 0.022 | |||
Observations | 2583 | |||
R2 / Ω02 | .452 / .442 |
Not a big difference, again. And if we remove the momentary effect?
interest | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 2.83 | 0.08 | <.001 | |
youth_activity_rc (Basic Skills Activity) | -0.00 | 0.06 | .974 | |
youth_activity_rc (Creating Product) | 0.13 | 0.06 | .037 | |
youth_activity_rc (Field Trip Speaker) | 0.13 | 0.12 | .262 | |
youth_activity_rc (Lab Activity) | 0.05 | 0.12 | .677 | |
youth_activity_rc (Program Staff Led) | 0.04 | 0.07 | .610 | |
Random Parts | ||||
Nbeep_ID_new | 235 | |||
Nparticipant_ID | 203 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.036 | |||
ICCparticipant_ID | 0.325 | |||
ICCprogram_ID | 0.021 | |||
Observations | 2826 | |||
R2 / Ω02 | .450 / .441 |
Let’s take a look at engagement as the outcome.
Same as model 1, but with engagement as the outcome.
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.89 | 0.20 | <.001 | |
challenge | 0.05 | 0.01 | <.001 | |
relevance | 0.54 | 0.02 | <.001 | |
gender (M) | -0.07 | 0.05 | .169 | |
classroom_versus_field_enrichment | 0.10 | 0.04 | .010 | |
CLASS_comp | 0.01 | 0.02 | .781 | |
youth_activity_rc (Basic Skills Activity) | -0.00 | 0.04 | .965 | |
youth_activity_rc (Creating Product) | -0.06 | 0.05 | .255 | |
youth_activity_rc (Field Trip Speaker) | 0.04 | 0.08 | .614 | |
youth_activity_rc (Lab Activity) | 0.05 | 0.08 | .528 | |
youth_activity_rc (Program Staff Led) | -0.07 | 0.05 | .172 | |
overall_pre_interest | 0.08 | 0.03 | .020 | |
youth_development_overall | 0.01 | 0.01 | .371 | |
prop_attend | 0.17 | 0.18 | .344 | |
Random Parts | ||||
Nbeep_ID_new | 227 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.054 | |||
ICCparticipant_ID | 0.253 | |||
ICCprogram_ID | 0.020 | |||
Observations | 2428 | |||
R2 / Ω02 | .687 / .685 |
This model seems to tell us:
What if we remove challenge and relevance?
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 2.00 | 0.32 | <.001 | |
gender (M) | 0.06 | 0.09 | .529 | |
classroom_versus_field_enrichment | 0.08 | 0.05 | .065 | |
CLASS_comp | 0.02 | 0.02 | .252 | |
youth_activity_rc (Basic Skills Activity) | 0.06 | 0.05 | .256 | |
youth_activity_rc (Creating Product) | 0.05 | 0.06 | .429 | |
youth_activity_rc (Field Trip Speaker) | 0.15 | 0.10 | .113 | |
youth_activity_rc (Lab Activity) | 0.08 | 0.09 | .397 | |
youth_activity_rc (Program Staff Led) | -0.01 | 0.06 | .857 | |
overall_pre_interest | 0.13 | 0.06 | .016 | |
youth_development_overall | 0.00 | 0.01 | .682 | |
prop_attend | 0.27 | 0.31 | .386 | |
Random Parts | ||||
Nbeep_ID_new | 227 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.031 | |||
ICCparticipant_ID | 0.439 | |||
ICCprogram_ID | 0.025 | |||
Observations | 2428 | |||
R2 / Ω02 | .550 / .545 |
About the same.
What if we remove the random beep effect from model 2?
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.90 | 0.20 | <.001 | |
challenge | 0.05 | 0.01 | <.001 | |
relevance | 0.54 | 0.02 | <.001 | |
gender (M) | -0.07 | 0.05 | .188 | |
classroom_versus_field_enrichment | 0.12 | 0.03 | <.001 | |
CLASS_comp | 0.00 | 0.01 | .736 | |
youth_activity_rc (Basic Skills Activity) | -0.00 | 0.03 | .918 | |
youth_activity_rc (Creating Product) | -0.08 | 0.04 | .038 | |
youth_activity_rc (Field Trip Speaker) | 0.04 | 0.06 | .436 | |
youth_activity_rc (Lab Activity) | 0.04 | 0.06 | .485 | |
youth_activity_rc (Program Staff Led) | -0.06 | 0.04 | .092 | |
overall_pre_interest | 0.08 | 0.03 | .022 | |
youth_development_overall | 0.00 | 0.00 | .358 | |
prop_attend | 0.17 | 0.18 | .349 | |
Random Parts | ||||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCparticipant_ID | 0.249 | |||
ICCprogram_ID | 0.030 | |||
Observations | 2428 | |||
R2 / Ω02 | .647 / .647 |
Some of the coefficients for momentary predicts change a bit (creating product is significantly less engaging than not focused); program staff led demonstrates a similar pattern.
Lets look at the same model but with challenge and relevance removed.
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 1.98 | 0.32 | <.001 | |
gender (M) | 0.07 | 0.09 | .473 | |
classroom_versus_field_enrichment | 0.10 | 0.04 | .006 | |
CLASS_comp | 0.03 | 0.02 | .096 | |
youth_activity_rc (Basic Skills Activity) | 0.06 | 0.04 | .123 | |
youth_activity_rc (Creating Product) | 0.02 | 0.05 | .602 | |
youth_activity_rc (Field Trip Speaker) | 0.16 | 0.07 | .020 | |
youth_activity_rc (Lab Activity) | 0.08 | 0.07 | .279 | |
youth_activity_rc (Program Staff Led) | -0.00 | 0.04 | .930 | |
overall_pre_interest | 0.14 | 0.06 | .016 | |
youth_development_overall | 0.00 | 0.01 | .735 | |
prop_attend | 0.27 | 0.31 | .388 | |
Random Parts | ||||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCparticipant_ID | 0.435 | |||
ICCprogram_ID | 0.033 | |||
Observations | 2428 | |||
R2 / Ω02 | .504 / .501 |
These seem a bit more sensible; field trip speaker is associated with higher engagement; otherwise, there are similar patterns.
We may be having too many predictors. What can we remove?
Let’s consider removing some of the highly not significant predictors: CLASS composite, youth development from PQA, maybe prop attend from model 2.
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 0.89 | 0.20 | <.001 | |
challenge | 0.05 | 0.01 | <.001 | |
relevance | 0.54 | 0.02 | <.001 | |
gender (M) | -0.07 | 0.05 | .169 | |
classroom_versus_field_enrichment | 0.10 | 0.04 | .010 | |
CLASS_comp | 0.01 | 0.02 | .781 | |
youth_activity_rc (Basic Skills Activity) | -0.00 | 0.04 | .965 | |
youth_activity_rc (Creating Product) | -0.06 | 0.05 | .255 | |
youth_activity_rc (Field Trip Speaker) | 0.04 | 0.08 | .614 | |
youth_activity_rc (Lab Activity) | 0.05 | 0.08 | .528 | |
youth_activity_rc (Program Staff Led) | -0.07 | 0.05 | .172 | |
overall_pre_interest | 0.08 | 0.03 | .020 | |
youth_development_overall | 0.01 | 0.01 | .371 | |
prop_attend | 0.17 | 0.18 | .344 | |
Random Parts | ||||
Nbeep_ID_new | 227 | |||
Nparticipant_ID | 176 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.054 | |||
ICCparticipant_ID | 0.253 | |||
ICCprogram_ID | 0.020 | |||
Observations | 2428 | |||
R2 / Ω02 | .687 / .685 |
These seem to change a bit, but not much. What if we remove challenge and relevance?
df$overall_engagement | ||||
B | std. Error | p | ||
Fixed Parts | ||||
(Intercept) | 2.33 | 0.18 | <.001 | |
gender (M) | 0.06 | 0.09 | .516 | |
classroom_versus_field_enrichment | 0.07 | 0.04 | .095 | |
youth_activity_rc (Basic Skills Activity) | 0.08 | 0.05 | .070 | |
youth_activity_rc (Creating Product) | 0.11 | 0.05 | .021 | |
youth_activity_rc (Field Trip Speaker) | 0.16 | 0.09 | .093 | |
youth_activity_rc (Lab Activity) | 0.12 | 0.09 | .167 | |
youth_activity_rc (Program Staff Led) | 0.01 | 0.05 | .862 | |
overall_pre_interest | 0.13 | 0.05 | .017 | |
Random Parts | ||||
Nbeep_ID_new | 235 | |||
Nparticipant_ID | 180 | |||
Nprogram_ID | 9 | |||
ICCbeep_ID_new | 0.032 | |||
ICCparticipant_ID | 0.440 | |||
ICCprogram_ID | 0.018 | |||
Observations | 2583 | |||
R2 / Ω02 | .546 / .541 |
If we remove challenge and relevance, creating product has a positive relation, and field trip speaker is approaching significance.