2024年5月1日发(作者：redmi k30 pro参数)

Stata常用指令

set more off

set virtual on

di exp(3.567)

解釋

= display

Browse the data

tabmiss x1 x2 (findit tabmiss)

browse var1 var2 (if ….)

顯示MV的freq與比例

Look like editor window, but

cannot edit

listblck in 1/10, repeat(1)

(findit listblck)

list, 但將版面精縮

repeat(1/n) => 前1(n)個重複出

現after row 2

(findit univar)

univar chinese math science, boxplot

, by(gender) onehdr

univar math, by(gender) onehdr boxplot onescal

 univar (=sum) 但增加

q25, midian, q75的呈現

 get a table with one header

 onescale才能相比較

Summary Statistics & Tables

sum

we can use if : eg. (if crime==1)

tab x1, sort miss

(sort=按照distribution排列; miss=列出MV distribution as well)

ta x1 x2, chi2 miss

, nof column (no frequency / column percentage)

, row (row percentage)

, all (all available statistics)

, exact (Fisher’s exact test)

ta maage_group, plot

tab1 x1 x2 x3 x4

tab2 x1 x2 x3 x4

ta paedu, sum(crime)

Summarize all variables

(mean, SD, freqency)

tab=tabulate

Chi2=Pearson chi-square test of

independence

= tab x1 / tab x2…….

tab all possible two-way..

By levels of paedu, summarize

crime

tabstat score, stats(mean sd n max min…) by (subject) median, p10, p25, iqr, q…

iqr=interquaritile range=p75-p25

q=quartiles= if we specify p25 p50 p75

table x1 x2, contents(mean y1 median y2)

Also min, max….etc…..

Data Management

gen id=_n (then do something else)

sort id

browse var1 var2 (if ….)

If want to come back to the

earlier order…..

Look like editor window, but

cannot edit

edit var1 var2 var3 (if…)

label variable bw “birth weight”

drop if id==id[_n-1] & birthday==birthday[_n-1]

Or just replace delete=1, 就不

用真的delete

format id %9.0f

encode region, gen(region2)

tab region2 (looks the same but…)

tab region2, nolabel (now we see the numeric value)

mvdecode

mvencode

egen zscore=std(x)

egen avg=rmean(Chinese, English, math)

egen sum=rsum(x,y,z)

list population region, nolabel

(only for lebeled numeric variables, not string var)

[分組]

sort var

gen varnew=group(5)

egen iseicat=cut(isei), at(10, 40, 70, 90)

table iseicat, contents(min isei max isei) => 檢查

字元太多不夠顯現時….

It generate labeled- numeric var

from a string variable.

numeric value => mv

mv => numeric value

標準分數 (mean=0, V=1)

Row mean, ignore MV

Row sum, MV=0

Display numeric var instead of

the labels

分成相同cases五組

分成10, 40, 70三組

不包括上限 (eg.90)

不被包括者 => MV

egen iseicat=cut(isei), at(10, 40, 70, 90) icodes

egen iseicat=cut(isei), at(10, 40, 70, 90) label

=> 變成 0, 1, 2 三組

=> 跟icodes一樣，但加了label

(10- 40- 70- )

local x "st2 st3 " [for later use: type `x']

定義長字串

Importing data from other programs

infile str30 place population sex score using

String var之前要加str#, as

many as #characters

(clean Excel data following stata data format)

(save Excel as .csv file )

insheet using “c:/data/”

infix

reshape?

collapse?

Excel => stata data

Compare groups

ttest college, by(male)

Regression

by region3, sort:reg score paedu

sort region3

by region3:reg score paedu

reg y x1 x2 x3, beta

sw reg Y x1 x2 x3 x4 x5….., pr(.05)

pr=p to retain (backward elimination)

sw reg Y x1 x2 x3 x4 x5….., pe(.05)

standardized regression

Stepwise reg:

它自己remove不顯著Xs

pe=p to enter

After regression…

predict yhat

predict e, risid

sort e

list v1 v2 v3… in 1/10 (or in -10/l) (l=last, not one)

lstat ?

Listcoef, help (要search & install: Long’s spostdo)

Residual

We can examine where the

model fits poorly…

correct classification rate

列出 X(&Y)的標準化係數

After logistic regression

est store full

quietly logistic y x (nested model)

lrtest full

logit y x

predict phat

graph twoway connected phat x, sort

predict q, xb

predict phat

graph twoway mspline phat x2

adjust, by(var1) exp 後者=前者*exp(b)

adjust, by(var1) pr p/(1-p)=odds (when var1=n)

Interaction term的詮釋: B1(Main)+B2(dummy)

For the group (dummy=1): the odds ratio of Main is

logistic y var1 var2 inter

lincom var1+inter

lincom [2]lbw+[2]inter10, or (for mlogit) ([2]=model)

Likelihood-ratio test :

=> Phat=predicted p

=exp(a+bx)/[1+exp(a+bx)]

=> xb = lg odd = ln(p/(1-p))

=> odds when var1=1,2,3..

=> p(y) when var1=1,2,3..

exp(B1) * exp(B2)

Get point estimation & CI of

coefficient combination

用方便的方式得到 predicted probability

prchange (findit spost)

prchange, fromto help (help: add 說明)

prtab

prtab, x(paedu=1 maedu=1) rest(min)

prgen isei, f(30) t(60) gen(ff) x(male=0)

prgen isei, f(30) t(60) gen(mm) x(male=1)

twoway (connected ffp1 ffx) (connected mmp1 mmx)

xi3: logit y i.x1*male

postgr3 male, by(x1) table (very useful for obtain p)

postgr3 isei, by(area) (連續變項也可以)

Changes in predicted

probability

Predicted probability in n*n

table

連續變項對y=1的影響 （於範

圍內自動取n[default=11]點來

計算p）

有interaction term時……

=> male effect 因x1類別而不

同

mlogit

mlogit y x1 x2, rrr nolog base(2)

(ref group=> y=2)

rrr=relative risk ratio (=OR)

Output

outreg using , nolabel replace (findit outreg) & install

outreg using , nolabel append

Then convert text into table

儲存時要click no另存新檔

append = model 2 add on M1

outreg var1 var2 using , replace 10pct coefastr se

(se=st. error instead of t statistics)

log using , replace (don’t use t)

最後：log2html , replace

可指定列出哪些係數

(+ p<.1) (* add on coef)

(先 findit log2html)

=> 可以把結果存成html

List all the graph files

i.e., is saved

test whether a mediator carries

the influence of an IV to a DV.

Shortcut=program name we set

=>shortcut 本身變成command

_merge=1 (from master data),

2=from using data…

Graph

graph dir

graph use gender_gap

graph save filename

erase

其他

sgmediation var_y, mv(varx1) iv(varx2)

[Sobel-Goodman tests: use findit first]

省時

program define shortcut

command 1 … command 2

end

shortcut (自己跑一遍command 1, 2..)

超級常用

list, gen, recode, replace, rename, sort, drop, keep, order……

merge, append