Dear Statalist,
My question concerns the correct wording when discussing results from regressions, margins, average marginal effects, and pwcompare.
In a simple OLS regression, I would say: the treatment condition (6 versus 3 years) is associated with an insignificant increase in action of 3.5 percentage points. Other variations are "increases the probability" and "participants after 6 years are more 3.5 pp more likely to be active compared to 3 years" [I know that I have insignificance here, I'm more worried about the correct wording].
Now, including the interaction between gender and treatment condition:
Now, with average marginal effects via -margins T_C, dydx(gender) after the interaction, where 1.gender is male and 2.gender is female: I would still say Women after 3 years are 14 percentage points less likely to perform an action compared to men after the same time frame. Is this correct?
Now my Q: do I need to say predicted in every sentence - Women after 3 years are predicted to be 14 percentage points less active than men after the same time frame? In Stata guides, the authors said the word predicted, but they only talked about one result (https://stats.oarc.ucla.edu/stata/se...actions-stata/), whereas I would like to make more than one comparison. I'm afraid all my sentences will be very hard to read, especially in my results section.
For completeness, the -pwcompare (asobserved) command, because I seek to understand what actually happened in my data compared to if the data were equal. This command also states that these are marginal linear predictions.
Of course, I will post proper results tables in the paper such that it will be evident how I arrived at the results.
My question concerns only the proper wording. In which situations can I talk about probability, and where do I need to state predictions explicitly? From the interaction regression, taking T_C 6 years and adding that to T_C#gender 6 years#female will give me the (female#6 years) vs (female#3 years) coefficient, which (to me) indicates that all these analyses are very interlinked and which also can be obtained via:
Thank you very much in advance!
My question concerns the correct wording when discussing results from regressions, margins, average marginal effects, and pwcompare.
In a simple OLS regression, I would say: the treatment condition (6 versus 3 years) is associated with an insignificant increase in action of 3.5 percentage points. Other variations are "increases the probability" and "participants after 6 years are more 3.5 pp more likely to be active compared to 3 years" [I know that I have insignificance here, I'm more worried about the correct wording].
Code:
Linear regression Number of obs = 2,701
F(9, 780) = 3.49
Prob > F = 0.0003
R-squared = 0.0189
Root MSE = .47867
(Std. err. adjusted for 781 clusters in CASE)
-------------------------------------------------------------------------------
| Robust
action | Coefficient std. err. t P>|t| [95% conf. interval]
--------------+----------------------------------------------------------------
T_C |
6 years | .034639 .0277881 1.25 0.213 -.0199094 .0891874
|
gender |
female | -.1441114 .0360066 -4.00 0.000 -.2147928 -.0734301
non-binary | -.2839578 .1198681 -2.37 0.018 -.5192601 -.0486554
not stated | .0824881 .1462172 0.56 0.573 -.2045377 .3695138
|
device |
Tablet | -.0058154 .1186563 -0.05 0.961 -.2387389 .2271082
Smartphone | -.0275835 .0384659 -0.72 0.474 -.1030924 .0479255
|
system |
Android | .0211903 .0458835 0.46 0.644 -.0688795 .1112601
Apple | .0280029 .1449563 0.19 0.847 -.2565478 .3125536
|
1.instruclick | .0801797 .0296241 2.71 0.007 .0220274 .1383321
_cons | .3630647 .0252171 14.40 0.000 .3135633 .4125661
-------------------------------------------------------------------------------
Now, including the interaction between gender and treatment condition:
Code:
Linear regression Number of obs = 2,701
F(11, 780) = .
Prob > F = .
R-squared = 0.0196
Root MSE = .47876
(Std. err. adjusted for 781 clusters in CASE)
-------------------------------------------------------------------------------------
| Robust
action | Coefficient std. err. t P>|t| [95% conf. interval]
--------------------+----------------------------------------------------------------
T_C |
6 years | .0364919 .0309036 1.18 0.238 -.0241722 .097156
|
gender |
female | -.1440245 .0463511 -3.11 0.002 -.2350122 -.0530368
non-binary | .1276467 .0384864 3.32 0.001 .0520976 .2031958
not stated | .1025048 .1851297 0.55 0.580 -.2609067 .4659163
|
T_C#gender |
6 years#female | -.000118 .0725357 -0.00 0.999 -.1425063 .1422702
6 years#non-binary | -.5328814 .0692998 -7.69 0.000 -.6689177 -.3968451
6 years#not stated | -.0520135 .3003779 -0.17 0.863 -.6416584 .5376314
|
device |
Tablet | -.0019922 .1183719 -0.02 0.987 -.2343574 .2303729
Smartphone | -.0295926 .0386401 -0.77 0.444 -.1054436 .0462583
|
system |
Android | .0241251 .0460757 0.52 0.601 -.0663219 .1145721
Apple | .0238317 .1445695 0.16 0.869 -.2599597 .3076231
|
1.instruclick | .0791287 .029652 2.67 0.008 .0209215 .137336
_cons | .3623816 .0261706 13.85 0.000 .3110084 .4137548
-------------------------------------------------------------------------------------
Now, with average marginal effects via -margins T_C, dydx(gender) after the interaction, where 1.gender is male and 2.gender is female: I would still say Women after 3 years are 14 percentage points less likely to perform an action compared to men after the same time frame. Is this correct?
Now my Q: do I need to say predicted in every sentence - Women after 3 years are predicted to be 14 percentage points less active than men after the same time frame? In Stata guides, the authors said the word predicted, but they only talked about one result (https://stats.oarc.ucla.edu/stata/se...actions-stata/), whereas I would like to make more than one comparison. I'm afraid all my sentences will be very hard to read, especially in my results section.
Code:
Average marginal effects Number of obs = 2,701
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: 2.gender 3.gender 4.gender
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
1.gender | (base outcome)
-------------+----------------------------------------------------------------
2.gender |
T_C |
3 years | -.1440245 .0463511 -3.11 0.002 -.2350122 -.0530368
6 years | -.1441426 .0561392 -2.57 0.010 -.2543444 -.0339407
-------------+----------------------------------------------------------------
3.gender |
T_C |
3 years | .1276467 .0384864 3.32 0.001 .0520976 .2031958
6 years | -.4052347 .0530736 -7.64 0.000 -.5094188 -.3010507
-------------+----------------------------------------------------------------
4.gender |
T_C |
3 years | .1025048 .1851297 0.55 0.580 -.2609067 .4659163
6 years | .0504913 .2372121 0.21 0.831 -.4151584 .516141
------------------------------------------------------------------------------
Code:
Pairwise comparisons of marginal linear predictions
Margins: asobserved
-------------------------------------------------------------------------------------------------------
| Unadjusted Unadjusted
| Contrast Std. err. t P>|t| [95% conf. interval]
--------------------------------------+----------------------------------------------------------------
gender#T_C |
(male#6 years) vs (male#3 years) | .0364919 .0309036 1.18 0.238 -.0241722 .097156
(female#3 years) vs (male#3 years) | -.1440245 .0463511 -3.11 0.002 -.2350122 -.0530368
(female#6 years) vs (male#3 years) | -.1076507 .0562544 -1.91 0.056 -.2180786 .0027773
(female#3 years) vs (male#6 years) | -.1805164 .0459586 -3.93 0.000 -.2707336 -.0902992
(female#6 years) vs (male#6 years) | -.1441426 .0561392 -2.57 0.010 -.2543444 -.0339407
(female#6 years) vs (female#3 years) | .0363739 .0654454 0.56 0.579 -.0920961 .1648438
-------------------------------------------------------------------------------------------------------
My question concerns only the proper wording. In which situations can I talk about probability, and where do I need to state predictions explicitly? From the interaction regression, taking T_C 6 years and adding that to T_C#gender 6 years#female will give me the (female#6 years) vs (female#3 years) coefficient, which (to me) indicates that all these analyses are very interlinked and which also can be obtained via:
Code:
Average marginal effects Number of obs = 2,701
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: 1.T_C
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
0.T_C | (base outcome)
-------------+----------------------------------------------------------------
1.T_C |
gender |
male | .0364919 .0309036 1.18 0.238 -.0241722 .097156
female | .0363739 .0654454 0.56 0.579 -.0920961 .1648438
non-binary | -.4963895 .0628739 -7.89 0.000 -.6198117 -.3729673
not stated | -.0155216 .298784 -0.05 0.959 -.6020374 .5709943
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
Comment