Hi everyone, I am using Stata version 14.2. I apologize in advance for the long post and questions.
I am running a regression with the dependent variable "willingness to pay for health care" which is on an ordinal scale 1-9 with 9 being willing to pay the most.
I also have an experiment group which received information about taxes which is expected to have less favorable views on willingness to pay (experiment4control) which is a binary variable.
I want to see how the experiment group interacts with a quality perception of health care (lika_sjuk) which answers to the question "is health care provided equally?" also on a 1-9 scale and income as well.
Some things I should clarify before I write the output.
I run a linear regression instead of ologit in order to get results easier to interpret. The quality perception and income are both on ordinal scales but I treat them as continuous (also because I think it will be easier to interpret). I'll take any suggestions/comments to improve the model.
Here the interaction of the experiment group with quality perception and income squared shows to be significant.
I use the margins command
I choose income 6-13 because it's on an ordinal scale and those who are *<6 are not of interest in the experiment. For the quality perception I chose 8 and 9 for this example because they are the ones who are most willing to pay for health care. 
Ok. Those who are in the experiment group and believe health care is provided equally to a large extent (the green and orange lines) are more willing to pay for health care if they earn between 26-30k and 37-45k when compared to the control group(dont like income as categories but it was coded like this).
Questions:
1) Is there a way to also graph the linear interaction? or is it not needed since the squared term is what matters here.
2) What if the linear interaction was not significant but only the squared term, what would this mean? When a quadratic term is introduced, should I disregard the lower order term?
3) How do I know if income squared is significant for both groups? Could it be the case that it is only significant for the control group but then when I graph it, stata uses the same coefficient for both groups?
4) How can I graph the average and average +1 standard deviation for the different groups?
I appreciate all comments and suggestions.
Thanks!
I am running a regression with the dependent variable "willingness to pay for health care" which is on an ordinal scale 1-9 with 9 being willing to pay the most.
I also have an experiment group which received information about taxes which is expected to have less favorable views on willingness to pay (experiment4control) which is a binary variable.
I want to see how the experiment group interacts with a quality perception of health care (lika_sjuk) which answers to the question "is health care provided equally?" also on a 1-9 scale and income as well.
Some things I should clarify before I write the output.
I run a linear regression instead of ologit in order to get results easier to interpret. The quality perception and income are both on ordinal scales but I treat them as continuous (also because I think it will be easier to interpret). I'll take any suggestions/comments to improve the model.
Code:
regress bet_sjuk experiment4control##c.lika_sjuk##c.income1##c.income1 private1 woman edu
soctrust poltrust if working1==1 & income1>=6
Source SS df MS Number of obs = 810
F(16, 793) = 10.28
Model 449.688103 16 28.1055064 Prob > F = 0.0000
Residual 2168.54029 793 2.73460314 R-squared = 0.1718
Adj R-squared = 0.1550
Total 2618.2284 809 3.23637626 Root MSE = 1.6537
bet_sjukvard Coef. Std. Err. t P>t [95% Conf. Interval]
experiment4control
experiment4 19.61751 7.623833 2.57 0.010 4.652236 34.58279
lika_sjuk 2.987437 1.17922 2.53 0.011 .6726748 5.302199
experiment4control#c.lika_sjuk
experiment4 -3.809897 1.46709 -2.60 0.010 -6.689736 -.9300588
income1 3.525733 1.295486 2.72 0.007 .982746 6.068721
experiment4control#c.income1
experiment4 -3.976378 1.617875 -2.46 0.014 -7.152202 -.8005547
c.lika_sjuk#c.income1 -.6732699 .2439314 -2.76 0.006 -1.152097 -.1944422
experiment4control#c.lika_sjuk#c.income1
experiment4 .8154774 .3065451 2.66 0.008 .2137416 1.417213
c.income1#c.income1 -.1986045 .0673548 -2.95 0.003 -.3308194 -.0663897
experiment4control#c.income1#c.income1
experiment4 .1986081 .084519 2.35 0.019 .0327007 .3645155
c.lika_sjuk#c.income1#c.income1 .0367185 .0124429 2.95 0.003 .0122937 .0611434
experiment4control#c.lika_sjuk#c.income1#c.income1
experiment4 -.0428043 .0157648 -2.72 0.007 -.0737499 -.0118588
private1 -.4625278 .1326368 -3.49 0.001 -.7228886 -.202167
woman .3049564 .1279261 2.38 0.017 .0538426 .5560703
edu -.019273 .0345162 -0.56 0.577 -.087027 .0484809
soctrust .0789291 .0327923 2.41 0.016 .0145592 .143299
poltrust -.5292019 .089113 -5.94 0.000 -.7041271 -.3542768
_cons -8.012086 6.145549 -1.30 0.193 -20.07555 4.051381
I use the margins command
Code:
. margins, at(experiment4control=(0 1) income1=(6(1)13) lika_sjuk=(8 9)) vsquish
Ok. Those who are in the experiment group and believe health care is provided equally to a large extent (the green and orange lines) are more willing to pay for health care if they earn between 26-30k and 37-45k when compared to the control group(dont like income as categories but it was coded like this).
Questions:
Code:
experiment4control#c.income1#c.income1
experiment4 .1986081 .084519 2.35 0.019 .0327007 .3645155
experiment4control#c.lika_sjuk#c.income1#c.income1
experiment4 -.0428043 .0157648 -2.72 0.007 -.0737499 -.0118588
2) What if the linear interaction was not significant but only the squared term, what would this mean? When a quadratic term is introduced, should I disregard the lower order term?
3) How do I know if income squared is significant for both groups? Could it be the case that it is only significant for the control group but then when I graph it, stata uses the same coefficient for both groups?
4) How can I graph the average and average +1 standard deviation for the different groups?
I appreciate all comments and suggestions.
Thanks!
