I've got three questions (at the bottom) on how to interpret the Fairlie decomposition, which I haven't seen previous posts touching on and really hope you could help me with. Here I'm comparing the odds of being food insecure between male-headed households and female-headed households. "female" is the gender dummy of interest. "male"="female"==0. Below is the codes and output, following the order:
1. by(male)
2. by(female)
3. by(male) pooled (female)
4. by(female) pooled(female)
************************************************** ************************************************** *************
1. by(male)
************************************************** ************************************************** *************
2. by(female)
************************************************** ************************************************** *************
3. by(male) pooled (female)
************************************************** ************************************************** *************
4. by(female) pooled(female)
************************************************** ************************************************** *************
I have three questions based on the results.
1. Does the significance matter?
I want to make the case that "job" and "numkids" are two main variables that explain significant portion of the "female" coefficient. However, neither of them came out significant in any of the models. "disabled" and "nohome" were significant, but they were not of my interest. So I'm wondering, at the end of the day, if I can still make a story out of "job" and "numkids" despite their lack of significance.
2. What does it mean to have drastically different results from applying males' coefficients on females' distribution versus applying the females' coefficients on males' distribution?
The model 1 by(male) and model 2 by (female) gave me very different results in terms of the percentage of "female" coefficient explained. For instance, the percentage explained by all variables from model 1 is 63% versus -5.6% from model 2. What does this difference suggest?
3. What does it mean to "use a pooled model as reference"?
Model 3 and 4 gave me slightly different results, which makes me wonder if they were still applying coefficients of males and females, respectively, instead of "pooled" coefficients?
I would greatly appreciate any help you could offer.
Fei
1. by(male)
2. by(female)
3. by(male) pooled (female)
4. by(female) pooled(female)
************************************************** ************************************************** *************
1. by(male)
Code:
set seed 1987
fairlie insecure (educ: educ_2 educ_3 educ_4 educ_5) (job: job_2 job_3) (disabled: disabled_2 disabled_3) (numkids: numkids__lst1 numkids__lst3 numkids__lst4) (youngest3yo: youngest3yo__lst2 youngest3yo__lst3) noassets nohome (controls: race1_2 race1_3 race1_4 age_2 age_3 age_4 noncitizen_2 noncitizen_3 metro maritalyrs_2 maritalyrs_3 maritalyrs_4 divlength_max) [pw=whfnwgt_lst], by(male) ro rep(1000)
Non-linear decomposition by male (G) Number of obs = 2,384
N of obs G=0 = 1332
N of obs G=0 = 1052
Pr(Y!=0|G=0) = .19250152
Pr(Y!=0|G=1) = .15040188
Difference = .04209963
Total explained = -.00235132
------------------------------------------------------------------------------
insecure | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
educ | .0021776 .0017951 1.21 0.225 -.0013408 .005696
job | .0032274 .0067549 0.48 0.633 -.0100119 .0164668
disabled | -.0032917 .0012606 -2.61 0.009 -.0057625 -.0008209
numkids | .0120253 .0149308 0.81 0.421 -.0172386 .0412892
youngest3yo | -.0124641 .0084466 -1.48 0.140 -.0290192 .0040909
noassets | .0013563 .001574 0.86 0.389 -.0017286 .0044412
nohome | .0051656 .0023594 2.19 0.029 .0005413 .00979
controls | -.0106727 .0069248 -1.54 0.123 -.0242452 .0028997
------------------------------------------------------------------------------
2. by(female)
Code:
set seed 1987
fairlie insecure (educ: educ_2 educ_3 educ_4 educ_5) (job: job_2 job_3) (disabled: disabled_2 disabled_3) (numkids: numkids__lst1 numkids__lst3 numkids__lst4) (youngest3yo: youngest3yo__lst2 youngest3yo__lst3) noassets nohome (controls: race1_2 race1_3 race1_4 age_2 age_3 age_4 noncitizen_2 noncitizen_3 metro maritalyrs_2 maritalyrs_3 maritalyrs_4 divlength_max) [pw=whfnwgt_lst], by(female) ro rep(1000)
Non-linear decomposition by female (G)
Number of obs = 2,384
N of obs G=0 = 1052
N of obs G=0 = 1332
Pr(Y!=0|G=0) = .15040188
Pr(Y!=0|G=1) = .19250152
Difference = -.04209963
Total explained = -.02650542
------------------------------------------------------------------------------
insecure | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
educ | -.0013155 .0022858 -0.58 0.565 -.0057955 .0031645
job | -.0136604 .0129202 -1.06 0.290 -.0389836 .0116627
disabled | .00123 .0017205 0.71 0.475 -.0021421 .0046022
numkids | -.0044622 .0164155 -0.27 0.786 -.036636 .0277116
youngest3yo | .0007097 .013537 0.05 0.958 -.0258224 .0272418
noassets | -.002566 .001975 -1.30 0.194 -.006437 .001305
nohome | -.0016142 .0023172 -0.70 0.486 -.0061557 .0029274
controls | -.0049137 .0057762 -0.85 0.395 -.0162348 .0064074
------------------------------------------------------------------------------
3. by(male) pooled (female)
Code:
set seed 1987
fairlie insecure (educ: educ_2 educ_3 educ_4 educ_5) (job: job_2 job_3) (disabled: disabled_2 disabled_3) (numkids: numkids__lst1 numkids__lst3 numkids__lst4) (youngest3yo: youngest3yo__lst2 youngest3yo__lst3) noassets nohome (controls: race1_2 race1_3 race1_4 age_2 age_3 age_4 noncitizen_2 noncitizen_3 metro maritalyrs_2 maritalyrs_3 maritalyrs_4 divlength_max) [pw=whfnwgt_lst], by(male) pooled(female) ro rep(1000)
Non-linear decomposition by male (G) Number of obs = 2,384
N of obs G=0 = 1332
N of obs G=0 = 1052
Pr(Y!=0|G=0) = .19250152
Pr(Y!=0|G=1) = .15040188
Difference = .04209963
Total explained = .00925427
------------------------------------------------------------------------------
insecure | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
educ | .0014597 .0011047 1.32 0.186 -.0007054 .0036248
job | .005008 .0051372 0.97 0.330 -.0050608 .0150768
disabled | -.0023381 .0009436 -2.48 0.013 -.0041875 -.0004888
numkids | .0109413 .0095888 1.14 0.254 -.0078525 .0297351
youngest3yo | -.0098119 .0062634 -1.57 0.117 -.022088 .0024642
noassets | .0014328 .0010059 1.42 0.154 -.0005386 .0034043
nohome | .0030665 .0014327 2.14 0.032 .0002585 .0058745
controls | -.0006015 .0036101 -0.17 0.868 -.0076772 .0064741
------------------------------------------------------------------------------
4. by(female) pooled(female)
Code:
set seed 1987
fairlie insecure (educ: educ_2 educ_3 educ_4 educ_5) (job: job_2 job_3) (disabled: disabled_2 disabled_3) (numkids: numkids__lst1 numkids__lst3 numkids__lst4) (youngest3yo: youngest3yo__lst2 youngest3yo__lst3) noassets nohome (controls: race1_2 race1_3 race1_4 age_2 age_3 age_4 noncitizen_2 noncitizen_3 metro maritalyrs_2 maritalyrs_3 maritalyrs_4 divlength_max) [pw=whfnwgt_lst], by(female) pooled(female) ro rep(1000)
Non-linear decomposition by female (G)
Number of obs = 2,384
N of obs G=0 = 1052
N of obs G=0 = 1332
Pr(Y!=0|G=0) = .15040188
Pr(Y!=0|G=1) = .19250152
Difference = -.04209963
Total explained = -.00925427
------------------------------------------------------------------------------
insecure | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
educ | -.0011715 .001069 -1.10 0.273 -.0032667 .0009236
job | -.0050534 .0051603 -0.98 0.327 -.0151673 .0050605
disabled | .0022909 .0009474 2.42 0.016 .0004339 .0041478
numkids | -.011533 .0097453 -1.18 0.237 -.0306335 .0075675
youngest3yo | .0100584 .0064518 1.56 0.119 -.0025869 .0227037
noassets | -.0014394 .0010166 -1.42 0.157 -.003432 .0005531
nohome | -.0030138 .001422 -2.12 0.034 -.0058008 -.0002269
controls | .000504 .0036118 0.14 0.889 -.006575 .007583
------------------------------------------------------------------------------
I have three questions based on the results.
1. Does the significance matter?
I want to make the case that "job" and "numkids" are two main variables that explain significant portion of the "female" coefficient. However, neither of them came out significant in any of the models. "disabled" and "nohome" were significant, but they were not of my interest. So I'm wondering, at the end of the day, if I can still make a story out of "job" and "numkids" despite their lack of significance.
2. What does it mean to have drastically different results from applying males' coefficients on females' distribution versus applying the females' coefficients on males' distribution?
The model 1 by(male) and model 2 by (female) gave me very different results in terms of the percentage of "female" coefficient explained. For instance, the percentage explained by all variables from model 1 is 63% versus -5.6% from model 2. What does this difference suggest?
3. What does it mean to "use a pooled model as reference"?
Model 3 and 4 gave me slightly different results, which makes me wonder if they were still applying coefficients of males and females, respectively, instead of "pooled" coefficients?
I would greatly appreciate any help you could offer.
Fei
