The panelview package has three main functionalities:
(1) it plots the treatment status and missing values in a panel dataset;
(2) it visualizes variables of interest in a time-series fashion;
(3) it depicts the bivariate relationships between a treatment variable and an outcome variable either by unit or in aggregate.
These tools can help researchers better understand their panel data before conducting statistical analysis.
Date: Sep 24, 2023
Version: 1.0.0 (Github); 1.0.0 (Stata Statistical Software Components (SSC) archive)
Authors: Hongyu Mou, Yiqing Xu
Reference: Hongyu Mou, Licheng Liu, and Yiqing Xu. "Panel Data Visualization in R (panelView) and Stata (panelview)." Journal of Statistical Software, 107(7), 1-20, 2023. doi:10.18637/jss.v107.i07.
Update in v.1.0.0:
- Updated formal citation for the Journal of Statistical Software (JSS).
- Officially released version 1.0.0.
Please report bugs to [email protected] or [email protected].
We thank Yifan Huang for an early version of the program and Ziyi Liu for helpful suggestions.
Table of Contents
[TOC]
Firstly, users need to install several dependencies: grc1leg, gr0075, labutil, and sencode.
net install grc1leg, from(http://www.stata.com/users/vwiggins) replace
net install gr0075, from(http://www.stata-journal.com/software/sj18-4) replace
ssc install labutil, replace
ssc install sencode, replace
Then to install the panelview package with Stata version 15.1 or greater, one way is by typing ssc install:
cd "your_full_local_path_to_ado_folder"
ssc install panelview, all
Another way is using net install to install the up-to-date version of panelview:
cap ado uninstall panelview //in-case already installed
net install panelview, all replace from("https://yiqingxu.org/packages/panelview_stata")
An overview of the syntax is below. Note that Y, D, and X in the table are simply labels; they can represent any variable in a panel dataset.
| Sata.formula | Option | Type | Function |
|---|---|---|---|
| Y D X | missing |
show missingness in Y, D, or X. | |
| Y | missing |
show missingness in Y. | |
| Y D X | treat |
show treatment status of D given complete data in Y, D, and X. | |
| D X | ignoreY |
treat |
show treatment status of D given complete data in D and X. |
| D | ignoreY (default) |
treat |
show treatment status of D. |
| Y D X | outcome |
show time-series of Y colored in D given complete data in Y, D, and X. | |
| Y | outcome |
show time-series of Y. | |
| Y D X | bivariate |
show relationship of Y and D given complete data in Y, D, and X. |
The general syntax of the package can be summarized as:
panelview Y [D X] [if] [in], ///
i(varname) t(varname numeric) ///
type(string) ///
[ ///
continuoustreat ///
discreteoutcome ///
bytiming ///
ignoretreat ///
ignoreY ///
MYCOLor(string) ///
PREpost ///
xlabdist(integer 1) ///
ylabdist(integer 1) ///
bygroup ///
style(string) ///
byunit ///
theme(string) ///
lwd(string) ///
leavegap ///
bygroupside ///
displayall ///
bycohort ///
collapsehistory ///
* ///
]
where the subcommand can be:
| Subcommand | Description |
|---|---|
Y D X |
varlist of outcome variable, treatment variable, and covariates, respectively. Including covariates may change the look of the plot due to missing values in these covariates. |
if and in |
We recommend users to add variable that is not included in the varlist or i() / t() but appears in the if/ in subcommand to the varlist following panelview command. |
i() and t() |
Specify the unit (group) and time indicators. |
type() |
Use type(treat) to plot treatment assignment using a heatmap. Use type(outcome) to plot an outcome variable---or any variable---in a time series fashion. Use type(bivar) or type(bivariate) to plot the outcome and treatment variables against time in the same graph. Use type(miss) or type(missing) to plot the missing data status of a variable. |
continuoustreat |
The treatment variable is presented as a continuous variable. |
discreteoutcome |
When a variable is discrete, make sure panelview respects its discreteness in type(outcome) plots. |
bytiming |
Sort units by the timing of first receiving the treatment; if the timing is the same, then by the total number of periods exposed to the treatment. |
ignoretreat |
Omit the treatment indicator, that is, any variables after Y will be interpreted as covariates. |
ignoreY |
Show treatment status of the first variable in the varlist instead of the second (e.g., D in formula is D X, instead of X). It needs to be combined with type(treat) or type(missing) . If there is only one variable in the varlist, the option is turned on by default. |
MYCOLor() |
Change the color schemes; click here for sequential colors (3-9 colors). |
PREpost |
Distinguish the pre- and post-treatment periods for treated units. |
xlabdist() and ylabdist() |
Change integer gaps between labels on the x- and y-axes. Default is 1. |
bygroup |
Put each unit into different treatment groups, then plot them separately in a column when type(outcome) is invoked. |
style() |
Determine the style of the elements in a plot. The first and second entries define the style of the outcome and treatment, respectively. connected or c for connected lines, line or l for lines, bar or b for bars. |
byunit |
Plot the outcome and treatment variables against time by each unit when type(bivar) is invoked. |
theme(bw) |
Use the black and white theme (default when specified type(bivar)). |
lwd() |
Set the line width in type(bivar) (default is medium). |
leavegap |
Keep the time gap as a white bar if time is not evenly distributed (possibly due to missing data). |
bygroupside |
Arrange subfigures of bygroup in a row rather than in a column. |
displayall |
Show all units if the number of units is more than 500, otherwise we randomly select 500 units to present. |
bycohort |
Plot the average outcome lines based on unique treatment history. |
collapsehistory |
Plot only the unique treatment histories, including figures alongside the plot for the number of units whose histories are characterized by each pattern. |
First, we show how to visualize the dichotomous treatment in a panel dataset. The treatment may switch on and off or have missing values.
Using the turnout dataset (a balanced panel), we show the treatment status of Election Day Registration (EDR) in each state in a given year (Xu 2017). We can use the title option to change the title of the plot and change the titles of x- and y-axes through xtitle and ytitle, respectively. For DID-type panel data with a dichotomous treatment indicator, we can distinguish the pre- and post-treatment periods for treated units by specifying prepost.
In the plot below, turnout is the outcome, policy_edr is the treatment, policy_mail_in and policy_motor are covariates. Including covariates may change the plot because of missing values in these covariates.
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) xtitle("Year") ytitle("State") title("Treatment Status")
We can use the bytiming option to sort units by the timing of receiving the treatment (then by the total number of periods exposed to the treatment). We also use legend to change labels in the legend:
*bytiming
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) xtitle("Year") ytitle("State") title("Treatment Status") bytiming legend(label(1 "No EDR") label(2 "EDR"))
Distinguish the pre- and post-treatment periods for treated units by specifying prepost:
*prepost
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) xtitle("Year") ytitle("State") title("Treatment Status") prepost
Again, sort units by the timing of receiving the treatment:
*bytiming
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) xtitle("Year") ytitle("State") title("Treatment Status") prepost bytiming
Remove the labels on the y-axis by specifying ylabel("") or ylabel(none):
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) title("EDR Reform") ylabel("")
Change the color schemes for the controls and treated using the mycolor option. For example, PuBu indicates light purple to blue. Click here for more sequential colors' choice.
*mycolor(PuBu)
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) xtitle("Year") ytitle("State") title("Treatment Status") mycolor(PuBu) bytiming
If time is not evenly distributed, we can use leavegap to keep the time gap as an white bar. Otherwise, we will skip the time gap with an warning "Time is not evenly distributed (possibly due to missing data)."
*leavegap
drop if year==1924
drop if year==1928
drop if year==1940
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(treat) leavegap
For a panel dataset in which the treatment may switch on and off, we no longer differentiate between pre- and post-treatment statuses. To demonstrate how panelview can be used in a more general setting, the following plot uses the capacity dataset, which is used to investigate the effect of democracy, the treatment, on state capacity, the outcome (Wang and Xu 2018). demo is a binary indicator of regime type. From the figure below, we see quite a few cases of democratic reversals and that there are many missing values (the white area). We use the xlabdist and ylabdist option to change the gaps between labels on the x- and y-axes:
use capacity.dta, clear
panelview lnpop demo lngdp , i(country) t(year) type(treat) mycolor(Reds) title("Democracy and State Capacity") xlabdist(3) ylabdist(10)
If the varlist formula is D X, instead of Y D X, we can use ignoreY to show treatment status of D, which do not take the missing status of Y into consideration:
*ignoreY
panelview demo lngdp , i(country) t(year) type(treat) mycolor(Reds) title("Democracy and State Capacity") xlabdist(3) ylabdist(10) ignoreY
Sorting units based on the first period a unit receives the treatment gives a more appealing visual:
*bytiming
panelview lnpop demo lngdp, i(country) t(year) type(treat) mycolor(Reds) title("Democracy and State Capacity") xlabdist(3) ylabdist(10) bytiming
If the number of units is too much, we can use collapsehistory to plot only the unique treatment histories, including figures alongside the plot for the number of units whose histories are characterized by each pattern:
*collapsehistory
panelview lnpop demo lngdp, i(country) t(year) type(treat) xlabdist(3) collapsehistory title("Unique Treatment Histories")
Instead of indicate country as units, we use i(ccode) to indicate country code as units, which will change the label and sequence in our figure:
panelview lnpop demo lngdp, i(ccode) t(year) type(treat) mycolor(PuBu) title("Democracy and State Capacity") xlabdist(3) ylabdist(10)
Sort units based on the first period a unit receives the treatment and use ylabel(none) to remove the labels on the y-axis:
*bytiming
panelview lnpop demo lngdp, i(ccode) t(year) type(treat) mycolor(PuBu) title("Democracy and State Capacity: Treatement Status", size(medsmall)) bytiming xlabdist(3) ylabel(none)
Sometimes a dataset has many units and we only want to take a peak of a subset of the units. panelview allows users to specify the units to be shown by the if subcommand. Note that if any variable not included in the varlist or i() / t() following panelview appears in the if or in command, we recommend researchers to add such variable into the varlist following panelview. In the following figure, we plot the treatment statuses of the first 25 units:
use capacity.dta, clear
egen ccodeid = group(ccode)
panelview lnpop demo lngdp ccodeid if ccodeid >= 1 & ccodeid <= 26, i(ccode) t(year) type(treat) mycolor(PuBu) title("Democracy and State Capacity") xlabdist(3)
Sort units based on the first period a unit receives the treatment:
*bytiming
panelview lnpop demo lngdp ccodeid if ccodeid >= 26 & ccodeid <= 51, i(ccode) t(year) type(treat) mycolor(PuBu) title("Democracy and State Capacity") xlabdist(3) bytiming
When we omit the treatment variable in a type(treat) plot, the plot will show missing (the white area) and non-missing values only. Another way to achieve this goal is to set type(missing).
use capacity.dta, clear
panelview demo, i(ccode) t(year) type(treat) mycolor(Reds) title("Missing Values") xlabel(none) ylabel(none) ignoretreat
We can also combine with leavegap to plot an vertical white bar if time is not evenly distributed (possibly due to missing data):
*leavegap
replace demo=. if year==1960
replace demo=. if year==1980
replace lngdp=. if year==1990
panelview demo lngdp, i(ccode) t(year) type(missing) mycolor(Reds) leavegap ylabel(none)
If the treatment indicator has only 1 level, then treatment status will not be shown in the type(treat) plot, which is the same as ignoretreat:
use capacity.dta, clear
gen demo2 = 0
panelview Capacity demo2 lngdp, i(ccode) t(year) type(treat) title("Regime Type") xlabdist(3) ylabdist(10) legend(off) // type(treat) & number of treatment level = 1: same as ignoretreat
If the treatment indicator has only 1 level, then treatment status will not be shown in the type(outcome) plot, which is the same as ignoretreat:
use capacity.dta, clear
gen demo2 = 0
panelview Capacity demo2 lngdp, i(ccode) t(year) type(outcome) title("Regime Type") legend(off) // type(outcome) & number of treatment level = 1: same as ignoretreat
If the treatment indicator has more than 2 treatment levels or is a continuous variable, then treatment status will not be shown in the type(outcome) plot. In other words, type(outcome) combined with continuoustreat or > 2 treatment levels is the same as ignoretreat.
With a continuous treatment variable (e.g. polity2), the treatment status will not be shown on the type(outcome) plot. We also indicate theme(bw) for black and white color style.
use capacity.dta, clear
* Continuous Outcome: Capacity; Continuous Treatment: polity2
panelview Capacity polity2 lngdp, i(ccode) t(year) type(outcome) title("Measuring State Capacity") legend(off) theme(bw)
Same as the following two commands:
use capacity.dta, clear
panelview Capacity demo lngdp, i(ccode) t(year) type(outcome) title("Measuring State Capacity") ignoretreat legend(off)
* Treatment indicator has more than 2 treatment levels
* Continuous Outcome: Capacity
use capacity.dta, clear
gen demo2 = 0
replace demo2 = -1 if polity2 < -0.5
replace demo2 = 1 if polity2 > 0.5
tab demo2, m
panelview Capacity demo2 lngdp, i(ccode) t(year) type(outcome) title("Measuring State Capacity") legend(off) // number of treatment level = 3
When the number of treatment levels is more than two, the treatment status will not be shown on the type(outcome) plot:
use simdata.dta, replace
replace D = 2 if time < 5
tab D, m
panelview Y D, type(outcome) i(id) t(time) mycolor(Greens) discreteoutcome title("Raw Data") // number of treatment level = 3
Same as the following two commands:
use simdata.dta, replace
panelview Y D, type(outcome) i(id) t(time) mycolor(Greens) discreteoutcome title("Raw Data") ignoretreat
use simdata.dta, replace
range x 0 1
panelview Y x, type(outcome) i(id) t(time) discreteoutcome title("Raw Data") theme(bw) // continuous treatment & black and white theme
panelview supports panel data with more than 2 treatment levels. For example, we create a measure of regime type with three treatment levels:
use capacity.dta, clear
gen demo2 = 0
replace demo2 = -1 if polity2 < -0.5
replace demo2 = 1 if polity2 > 0.5
panelview Capacity demo2 lngdp, i(ccode) t(year) type(treat) title("Regime Type") xlabdist(3) ylabdist(10) mycolor(Reds) // type(treat) & number of treatment level = 3
use capacity.dta, clear
gen demo2 = 0
replace demo2 = -2 if polity2 < -0.7
replace demo2 = -1 if polity2 < -0.5 & polity2 > -0.7
replace demo2 = 1 if polity2 > 0.5
panelview Capacity demo2 lngdp, i(ccode) t(year) type(treat) title("Regime Type") xlabdist(3) ylabdist(10) mycolor(Reds) // number of treatment level = 4
If the number of treatment levels is greater than 5, then the treatment indicator will be regarded as a continuous variable.
use capacity.dta, clear
gen demo2 = 0
replace demo2 = -2 if polity2 < -0.7
replace demo2 = -1 if polity2 < -0.5 & polity2 > -0.7
replace demo2 = 1 if polity2 > 0.5 & polity2 < 0.7
replace demo2 = 2 if polity2 > 0.7
tab demo2, m
panelview Capacity demo2 lngdp, i(ccode) t(year) type(treat) title("Regime Type") xlabdist(3) ylabdist(10)
To plot the continuous treatment variable, we convert the continuous treatment variable into five groups according to its treatment levels.
In the following example, polity2 ranges from -1 to 1. When polity2 is in the lowest category (-1), it ranges from -1 to (but not including) -0.5. When polity2 is in the -0.5 category, it ranges from -0.5 to (but not including) 0. When polity2 is in the 0 category, it ranges from 0 to (but not including) 0.5. When polity2 is in the 0.5 category, it ranges from 0.5 to (but not including) 1. When polity2 is in the 1 category, it indicates observations when polity2 is equal to 1.
use capacity.dta, clear
panelview lngdp polity2, i(ccode) t(year) type(treat) mycolor(Reds) title("Regime Type") xlabdist(3) ylabdist(10)
If we change the level of the continuous treatment variable, the legend will modify correspondingly:
use capacity.dta, clear
replace polity2 = polity2 + 1
panelview lngdp polity2, i(ccode) t(year) type(treat) mycolor(Reds) title("Regime Type") xlabdist(3) ylabdist(10)
The second functionality of panelview is to show the raw outcome variable of a panel dataset in a time-series fashion. The syntax is very similar except that we need to specify type(outcome). Different colors represent different treatment conditions.
Note that we paint the period right before when the treatment begin as treated period. Different with type(treat), type(outcome)does not need xlabdist and ylabdist. If needed, we should use xlabel and ylabel instead.
* Continuous outcome: turnout: 0-100; Discrete Treatment: policy_edr: 0/1
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") title("EDR Reform and Turnout") ylabel(0 (25) 100)
Distinguish the pre- and post-treatment periods for treated units:
*prepost
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") title("EDR Reform and Turnout") prepost
Apply the light purple to blue theme by specifying mycolor(PuBu):
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") title("EDR Reform and Turnout") mycolor(PuBu)
We can specify which unit(s) we want to take a look at:
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor if abb == 1|abb == 2|abb == 6, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") title("EDR Reform and Turnout (AL, AR, CT)") mycolor(PuBu) prepost
To better understand the data, sometimes we want to plot the outcome based on whether the treatment status has changed during the observed time period. We can simply add options bygroup. The algorithm will analyze the data and automatically put each unit into different groups, e.g. (1) units always being treated, (2) units always under control, (3) units whose treatment status has changed.
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") by(, title("EDR Reform and Turnout")) bygroup xlabel(1920 (20) 2000)
If we want to arrange the subfigures in a row rather than in a column, use the bygroupside option instead of bygroup.
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") by(, title("EDR Reform and Turnout")) bygroupside xlabel(1920 (20) 2000)
Starting from v 0.1.3, users can use the bycohort option to plot the average outcome trajectories of units with same treatment history (if the number of unique treatment history is less than 20). This type of plots can help users diagnose the extent to which treatment effect heterogeneity may cause biases of certain estimators (e.g. the two-way fixed effects estimators).
use turnout.dta, clear
panelview turnout policy_edr, i(abb) t(year) type(outcome) bycohort prepost
We can accommodate discrete variables by setting discreteoutcome. Below is an example using the simdata dataset, in which the outcome variable takes three values: 0, 1, and 2.
use simdata.dta, replace
panelview Y D if time >= 8 & time <= 15, type(outcome) i(id) t(time) mycolor(Reds) discreteoutcome title("Raw Data") xlabel(8 (2) 15) ylabel(0 (1) 2)
We split the sample based on changes in treatment status:
use simdata.dta, replace
panelview Y D if time >= 8 & time <= 15, type(outcome) i(id) t(time) discreteoutcome by(,title("Raw Data")) xlabel(8 (2) 15) ylabel(0 (1) 2) bygroup
Plot an outcome variable (or any variable) in a panel dataset by type(outcome):
use turnout.dta, clear
panelview turnout, i(abb) t(year) type(outcome) xtitle("Year") ytitle("Turnout") title("Turnout") ylabel(0 (25) 100)
Visualize time series of the outcome and treatment in one figure by specifying type(bivar) or type(bivariate). For continuous variable, we use line plot as default; for discrete variable, we use bar plot. To plot connected lines ( connected or c ), lines ( line or l ), or bars ( bar or b ) rather than the default, please add style( , ), where the first element defines the outcome style, and the second defines the treatment style.
This section plots mean D and Y against time in the same graph.
With continuous outcome and discrete treatment, here are two examples. In the former one, style(c,b) means connected scatter instead of default line plot for the outcome and bar plot for the treatment. If any connected line, we can specify the symbol size by msize():
/***** 1. Y: continuous; D: discrete *****/
use turnout.dta, clear
*label the first and second y axes
panelview turnout policy_edr, i(abb) t(year) xlabdist(7) type(bivariate) msize(*0.5) style(c b) ytitle("turnout") ytitle("policy_edr", axis(2)) legend(label(1 "turnout") label(2 "policy_edr")) ylabel(40 (10) 70) ylabel(0 (0.1) 0.5, axis(2))
If not apply the default black and white theme, set option mycolor(). Besides, lwd(medthick) is to change the line width from the default medium to medthick:
use capacity.dta, clear
panelview lnpop demo, i(country) t(year) xlabdist(10) type(bivar) mycolor(Reds) lwd(medthick)
If the outcome is discrete, we can plot outcome and treatment against time in the same figure adding discreteoutcome:
/***** 2. Y: discrete; D: discrete *****/
use simdata.dta, replace
panelview Y D,i(id) t(time) discreteoutcome xlabdist(4) type(bivar) mycolor(Reds)
When treatment variable is continuous:
/***** 3. Y: continuous; D: continuous *****/
use capacity.dta, clear
panelview lnpop polity2, i(country) t(year) xlabdist(20) type(bivar)
In the last situation, we plot discrete outcome and continuous treatment:
/***** 4. Y: discrete; D: continuous *****/
use simdata.dta, replace
range x 0 1
panelview Y x, i(id) t(time) discreteoutcome xlabdist(4) type(bivar) style(b c)
We can add style(l,l) or style(line) to plot lines instead of bars for treatment:
/***** Line the discete treatment *****/
* Y: continuous; D: discrete
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor, i(abb) t(year) xlabdist(7) style(line) type(bivar)
*Y: discrete; D: discrete
use simdata.dta, replace
panelview Y D,i(id) t(time) discreteoutcome xlabdist(4) style(line) type(bivar)
We plot D and Y against time by each unit by option byunit. Below are two examples with continuous outcome and discrete treatment variable. We arrange four subgraphs in one row:
/***** 1. Y: continuous; D: discrete *****/
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor if abb >= 1 & abb <= 12, i(abb) t(year) xlabdist(10) type(bivar) byunit
use capacity.dta, clear
panelview lnpop demo if country >= 1 & country <= 24, i(country) t(year) xlabdist(20) type(bivar) byunit
With discrete outcome and treatment:
/***** 2. Y: discrete; D: discrete *****/
use simdata.dta, replace
panelview Y D if id >= 101 & id <= 120,i(id) t(time) discreteoutcome xlabdist(4) type(bivar) byunit
With continuous outcome and treatment:
/***** 3. Y: continuous; D: continuous *****/
use capacity.dta, clear
panelview lnpop polity2 if country >= 1 & country <= 12, i(country) t(year) xlabdist(20) type(bivar) byunit
With discrete outcome and continuous treatment:
/***** 4. Y: discrete; D: continuous *****/
use simdata.dta, replace
range x 0 1
panelview Y x if id >= 101 & id <= 112, i(id) t(time) discreteoutcome xlabdist(4) type(bivar) byunit
To better visualize a discrete treatment whose value is sometimes zero, add style(l l) to invoke line plots instead of bar plots:
/***** Line the discete treatment *****/
* Y: continuous; D: discrete
use turnout.dta, clear
panelview turnout policy_edr policy_mail_in policy_motor if abb >= 1 & abb <= 12, i(abb) t(year) xlabdist(10) style(l l) type(bivar) byunit
*Y: discrete; D: discrete
use simdata.dta, replace
panelview Y D if id >= 101 & id <= 120,i(id) t(time) discreteoutcome xlabdist(4) style(l l) type(bivar) byunit
