{smcl}
{* version 2.01 14mar2006}{...}
{cmd:help zerouse}
{hline}

{title:Title}

{p 4 21 2}
{hi:zerouse} {hline 2} Tabulate the pattern of zero digits in a variable

{title:Syntax}

{p 8 15 2}{cmdab:zerouse} {varname} [{cmd:,} {opth pat:tern(#.#)} {opth di:git(#)} {opth by:(varname)} {opth gen:erate(name)}
{it:{help zerouse##zerouse_tabulate_options:tabulate_options}}]

{title:Description}

{pstd}
The {cmd:zerouse} function displays the patterns of zero and non-zero digits
in a variable. The range of possible formats can be defined using an option,
as can the 'zero' digit, and results can be tabulated against a different 
variable or stored in a new variable for later use.

{title:Options}

{dlgtab:General}

{phang}
{opt pattern(#.#)} indicates the number of digits to the left and to the 
right of the decimal point that should be examined. The default is 9.9 and 
often this is a reasonable starting value although it may capture rounding errors.
Otherwise, select a value that will look at one digit more to the left and to
the right than you expect to find, and check that these extreme digits 
contain zeros rather than non-zero digits in the output.  You can also select a
pattern that is smaller than the digits used.  For example, to tabulate records
that do and do not have a zero in the first decimal place, use {opt pattern(0.1)}. 

{phang}
{opt by(varname)} displays results using a two way rather than one way frequency
table, giving you digit pattern cross-classified against your variable of choice.

{phang}
{opt digit(#)} is provided so you can use this program to investigate the pattern
of use of a digit other than zero.  For example, if you were to select the 
digit 2 then use of the digit 2 would be compared with the use of any digit 
other than 2 (including zero).

{phang}
{opt generate(name)} allows you to provide a name for a new variable into which
the patterns will be stored.  This means you can save the zero digit use pattern
for a number of variables and then cross tabulate them to identify consistent 
patterns of (for example) mistranscription.

{marker zerouse_tabulate_options}
{dlgtab:Tabulate Options}

{pstd}
Additional options can be passed to the {cmd:tabulate} command used to format output.
If the {opt by(varname)} option is not specified the output will be displayed 
in a one way table and the 
{help tabulate_oneway##tabulate1_options:tabulate oneway options} are available.
If the {opt by(varname)} option has been used, output will be displayed 
in a two way table and the 
{help tabulate_twoway##tabulate2_options:tabulate twoway options} can be used.

{title:Remarks}

{pstd}
This function is intended to assist with the identification of digit preferance 
problems when cleaning data, and to help identify any systematic relationship between
digit use and other variables, typically the laboratory or instrument that generated
the data or the clerk who transcribing the data.  It's use is most easily explained using an example:

    {cmd:. zerouse y1, pattern(2.1) sort}

          Digit {c |}
        pattern {c |}      Freq.     Percent        Cum.
    {dup 12:{c -}}{c +}{dup 38:{c -}}
           0#.0 {c |}     11,514       86.21       86.21
           ##.0 {c |}        851        6.37       92.59
           #0.0 {c |}        532        3.98       96.57
           0#.# {c |}        424        3.17       99.75
           00.0 {c |}         34        0.25      100.00
    {dup 12:{c -}}{c +}{dup 38:{c -}}
          Total {c |}     13,355      100.00

{pstd}
The example above shows zero digit use for the variable y1, from which the following can be gathered:

{pstd}
Most values are single digit integers (0#.0).  Some values have a tens digit of which about a third have a zero in the units position (#0.0) and two thirds have a digit in the units position (##.0).
This suggests the values are mainly below ten with a few values that are ten or above.

{pstd}
Notice that only a few values are not integers and these are all below ten (0#.#).  
These values may have been measured using greater precision but if this was the case we would expect at least a few values of ten or above (##.#).
Alternatively, it may be that some two digit values were incorrectly transcribed.  Perhaps a decimal point was inserted and this reduced these values by an order of magnitude.  Values with this digit 
pattern should probably be checked for transcription errors.  If there were multiple paths to the database (different data entry clerks, labs or instruments) and this 
is stored in the database it might also be useful to cross tabulate these digit formats with this information, to see if all the anomalous values originated in the same place.


{title:Examples}

    {cmd:. zerouse y1}
    {cmd:. zerouse y1, pattern(3.2) sort}
    {cmd:. zerouse y1, pattern(3.2) by(group)}
    {cmd:. zerouse y1, pattern(3.2) by(group) nofreq row}

    {cmd:. zerouse y1, pattern(3.2) digit(2)}

{title:Author}

{pstd}Richard J. Atkins{p_end}
{pstd}London School of Hygiene and Tropical Medicine{p_end}
{pstd}e-mail: richard.atkins@lshtm.ac.uk{p_end}