**[XT] xtdata** -- Faster specification searches with xt data

__Syntax__

**xtdata** [*varlist*] [*if*] [*in*] [**,** *options*]

*options* Description
-------------------------------------------------------------------------
Main
**re** convert data to a form suitable for random-effects
estimation
__r__**atio(***#***)** ratio of random effect to pure residual (standard
deviations)
**be** convert data to a form suitable for between estimation
**fe** convert data to a form suitable for fixed-effects
(within) estimation
**nodouble** keep original variable type; default is to recast type
as **double**
**clear** overwrite current data in memory
-------------------------------------------------------------------------
A panel variable must be specified; use **xtset**.

__Menu__

**Statistics > Longitudinal/panel data > Setup and utilities >** **Faster**
**specification searches with xt data**

__Description__

**xtdata** produces a transformed dataset of the variables specified in
*varlist* or of all the variables in the data. Once the data are
transformed, Stata's **regress** command may be used to perform specification
searches more quickly than **xtreg**; see **[R] regress** and **[XT] xtreg**. Using
**xtdata, re** also creates a variable named **constant**. When using **regress**
after **xtdata, re**, specify **noconstant** and include **constant** in the
regression. After **xtdata, be** and **xtdata, fe**, you need not include
**constant** or specify **regress**'s **noconstant** option.

__Options__

+------+
----+ Main +-------------------------------------------------------------

**re** specifies that the data be converted into a form suitable for
random-effects estimation. **re** is the default if **be**, **fe**, or **re** is not
specified. **ratio()** must also be specified.

**ratio(***#***)** (use with **xtdata, re** only) specifies the ratio sigma_u/sigma_e,
which is the ratio of the random effect to the pure residual. This
is the ratio of the standard deviations, not the variances.

**be** specifies that the data be converted into a form suitable for between
estimation.

**fe** specifies that the data be converted into a form suitable for
fixed-effects (within) estimation.

**nodouble** specifies that transformed variables keep their original types,
if possible. The default is to recast variables to **double**.

Remember that **xtdata** transforms variables to be differences from
group means, pseudodifferences from group means, or group means.
Specifying **nodouble** will decrease the size of the resulting dataset
but may introduce roundoff errors in these calculations.

**clear** specifies that the data may be converted even though the dataset
has changed since it was last saved on disk.

__Examples__

---------------------------------------------------------------------------
Setup
**. webuse xtdatasmpl**

Perform between transformation
**. xtdata ln_w grade age* ttl_exp* tenure* black not_smsa south, be**

Equivalent to **xtreg, be**
**. regress ln_w grade age* ttl_exp* tenure* black not_smsa south**

---------------------------------------------------------------------------
Setup
**. webuse xtdatasmpl, clear**

Perform within transformation
**. xtdata, fe**

Equivalent to **xtreg, fe**
**. regress ln_w grade age* ttl_exp* tenure* black not_smsa south**

---------------------------------------------------------------------------
Setup
**. webuse xtdatasmpl, clear**

Perform random-effects transformation
**. xtdata, re ratio(.88719358)**

Equivalent to **xtreg, re**
**. regress ln_w grade age* ttl_exp* tenure* black not_smsa south**
**constant, noconstant**
---------------------------------------------------------------------------

__Warnings__

1. **xtdata** is intended for use during the specification search phase of
analysis. Final results should be estimated with **xtreg** on
unconverted data.

2. Using **regress** after **xtdata, fe**, produces standard errors that are too
small, but only slightly.

3. Interpret significance tests and confidence intervals loosely. After
**xtdata, fe** and **re**, an incorrect (but close to correct) distribution
is being assumed.

4. You should ignore the summary statistics reported at the top of
**regress**'s output.

5. After converting the data, you may form linear, but not nonlinear
combinations of regressors.