Median Function in mySQL

[Ross Contino]

Hello to all:

I have been searching the web for examples on how to determine a median
value in a mySQL table. I have reviewed the article at
http://mysql.progen.com.tr/doc/en/Gr...functions.html. I am an
experienced VB programmer that has recently moved to PHP/mySQL. My employer
has a text file outputted from a vendor specific software with data.
However it cannot be manipulated because it is text. I created a web that
reads the text file on client, converts it to usable data, inputs to mySQL
table and then reflects HTML pages with desired report. One report
subtracts two times and then my employer wishes a median grouped by another
field. I used the AVG function, but of course this is mean, not Median. My
query is:

$mSql = "SELECT AdmitDoc,
Count(AdmitDoc) as Number_Admits,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Decdt))) as AVG_Doc_To_Admi t,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Indt))) as AVG_Total_Stay
from tadmits GROUP BY AdmitDoc";

This query subtracts several wait times in an emergency department and
groups by admitting physician. I would like to have median rather than
mean. AdmitDoc is text field, Outdt, Decdt, Indt are all date/time fields.

Any thoughts?

Thanks,
Ross

[Steve]

Ross Contino wrote:

Hello to all:

I have been searching the web for examples on how to determine a median value in a mySQL table. I have reviewed the article at
http://mysql.progen.com.tr/doc/en/Gr...functions.html. I am an
experienced VB programmer that has recently moved to PHP/mySQL. My employer has a text file outputted from a vendor specific software with data.
However it cannot be manipulated because it is text. I created a web that reads the text file on client, converts it to usable data, inputs to mySQL table and then reflects HTML pages with desired report. One report
subtracts two times and then my employer wishes a median grouped by another field. I used the AVG function, but of course this is mean, not Median. My query is:

$mSql = "SELECT AdmitDoc,
Count(AdmitDoc) as Number_Admits,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Decdt))) as AVG_Doc_To_Admi t,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Indt))) as AVG_Total_Stay
from tadmits GROUP BY AdmitDoc";

This query subtracts several wait times in an emergency department and groups by admitting physician. I would like to have median rather than mean. AdmitDoc is text field, Outdt, Decdt, Indt are all date/time fields.
Any thoughts?

Thanks,
Ross

There is no MEDIAN() operator in MySQL, so you will have to perform
this operation in several steps...

(1) Count the number of rows in each group
(2) For each group, if the count is odd, the median is the middle row
(when the set is ordered); when the count is even, the median is the
mean of the middle two rows.

In skeletal form...

// get counts for each set of interest...

SELECT AdmitDoc, COUNT(*) AS "Number_Adm its"
FROM tadmits
GROUP BY AdmitDoc

// for each AdmitDoc, work out LIMIT parameters (nb first row =
0)...

$doc = $datRow[ 'AdmitDoc' ];
$count = $datRow[ Number_Admits];

if $count is odd,
$limit1 = intval( $count / 2 );
$limit2 = 1;
else
$limit1 = intval( $count / 2 ) - 1;
$limit2 = 2;

// select middle row(s) from ORDERed set...

SELECT AdmitDoc,
SEC_TO_TIME(UNI X_TIMESTAMP(Out dt)-UNIX_TIMESTAMP( Decdt)) AS
"Doc_To_Adm it"
FROM tadmits
WHERE AdmitDoc = $doc
ORDER BY UNIX_TIMESTAMP( Outdt) - UNIX_TIMESTAMP( Decdt)
LIMIT $limit1, $limit2
if $count is odd,
// only one row...
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$median_doc_to_ admit = $datRow[ 'Doc_To_Admit' ];
else
// two rows
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val1 = $datRow[ 'Doc_To_Admit' ];
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val2 = $datRow[ 'Doc_To_Admit' ];
$median_doc_to_ admit = ( $val1 + $val2 ) / 2;

You'll have to do this separately for your two required measures
because they each need a different sort order.

---
Steve

[Steve]

Ross Contino wrote:

Hello to all:

I have been searching the web for examples on how to determine a median value in a mySQL table. I have reviewed the article at
http://mysql.progen.com.tr/doc/en/Gr...functions.html. I am an
experienced VB programmer that has recently moved to PHP/mySQL. My employer has a text file outputted from a vendor specific software with data.
However it cannot be manipulated because it is text. I created a web that reads the text file on client, converts it to usable data, inputs to mySQL table and then reflects HTML pages with desired report. One report
subtracts two times and then my employer wishes a median grouped by another field. I used the AVG function, but of course this is mean, not Median. My query is:

$mSql = "SELECT AdmitDoc,
Count(AdmitDoc) as Number_Admits,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Decdt))) as AVG_Doc_To_Admi t,
sec_to_time(AVG (unix_timestamp (Outdt) -
unix_timestamp( Indt))) as AVG_Total_Stay
from tadmits GROUP BY AdmitDoc";

This query subtracts several wait times in an emergency department and groups by admitting physician. I would like to have median rather than mean. AdmitDoc is text field, Outdt, Decdt, Indt are all date/time fields.
Any thoughts?

Thanks,
Ross

There is no MEDIAN() operator in MySQL, so you will have to perform
this operation in several steps...

(1) Count the number of rows in each group
(2) For each group, if the count is odd, the median is the middle row
(when the set is ordered); when the count is even, the median is the
mean of the middle two rows.

In skeletal form...

// get counts for each set of interest...

SELECT AdmitDoc, COUNT(*) AS "Number_Adm its"
FROM tadmits
GROUP BY AdmitDoc

// for each AdmitDoc, work out LIMIT parameters (nb first row =
0)...

$doc = $datRow[ 'AdmitDoc' ];
$count = $datRow[ Number_Admits];

if $count is odd,
$limit1 = intval( $count / 2 );
$limit2 = 1;
else
$limit1 = intval( $count / 2 ) - 1;
$limit2 = 2;

// select middle row(s) from ORDERed set...

SELECT AdmitDoc,
SEC_TO_TIME(UNI X_TIMESTAMP(Out dt)-UNIX_TIMESTAMP( Decdt)) AS
"Doc_To_Adm it"
FROM tadmits
WHERE AdmitDoc = $doc
ORDER BY UNIX_TIMESTAMP( Outdt) - UNIX_TIMESTAMP( Decdt)
LIMIT $limit1, $limit2
if $count is odd,
// only one row...
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$median_doc_to_ admit = $datRow[ 'Doc_To_Admit' ];
else
// two rows
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val1 = $datRow[ 'Doc_To_Admit' ];
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val2 = $datRow[ 'Doc_To_Admit' ];
$median_doc_to_ admit = ( $val1 + $val2 ) / 2;

You'll have to do this separately for your two required measures
because they each need a different sort order.

---
Steve

[Ross Contino]

Hi Steve:

I ended up writing a php function to select the medians from both values,
but it entailed looping the data frequently and creating a new table to
store it (see below). I could not quite get your method working. Would
your method have only worked via SQL and preclude the constant looping and
need for a table to store the medians?

Median functions:

/***********func tion block********** ***************/
function GetMedian_Dec($ myDoc, $No_Admits)
{
$MyMedian = "";
$LimitSql = "SELECT
SEC_TO_TIME(UNI X_TIMESTAMP(Out dt)-UNIX_TIMESTAMP( Decdt))AS Doc_To_Admit
FROM tadmits WHERE AdmitDoc = '" . $myDoc . "'" . "
ORDER BY Doc_To_Admit";
$result2 = mysql_query($Li mitSql);
$i=1;
$FirstTime = "";
$SecondTime = "";
If (($No_Admits % 2)==1){ //odd number of admissions
$MyCounter = intval(($No_Adm its / 2) + 1);
while($row3=mys ql_fetch_assoc( $result2)){
if ($i ==$MyCounter){
$UpSql = "UPDATE tersummary SET Median_MD_To_Ad mit = '"
.. $row3['Doc_To_Admit'] . "' WHERE AdmitDoc = '" . $myDoc . "'";
mysql_query($Up Sql);
}//end if
$i++;
}//end while
}
else { //even number of admissions
$MyCounter = intval($No_Admi ts / 2);
while($row3=mys ql_fetch_assoc( $result2)){
if ($i==$MyCounter ){
$FirstTime = $row3['Doc_To_Admit'];
}//end if
if ($i==$MyCounter + 1){
$SecondTime = $row3['Doc_To_Admit'];
}//end if
$i++;
}//end while
$AvgVal = (strtotime($Fir stTime) + strtotime($Seco ndTime)) /
2;
$UpSql = "UPDATE tersummary SET Median_MD_To_Ad mit = '" .
date("H:i:s",$A vgVal) . "' WHERE AdmitDoc = '" . $myDoc . "'";
mysql_query($Up Sql);
}
}
function GetMedian_Stay( $myDoc, $No_Admits)
{ $MyMedian = "";
$LimitSql = "SELECT
SEC_TO_TIME(UNI X_TIMESTAMP(Out dt)-UNIX_TIMESTAMP( Indt))AS Total_Stay
FROM tadmits WHERE AdmitDoc = '" . $myDoc . "'" . "
ORDER BY Total_Stay";
$result2 = mysql_query($Li mitSql);
$i=1;
$FirstTime = "";
$SecondTime = "";
If (($No_Admits % 2)==1){ //odd number of admissions
$MyCounter = intval(($No_Adm its / 2) + 1);
while($row3=mys ql_fetch_assoc( $result2)){
if ($i ==$MyCounter){
$UpSql = "UPDATE tersummary SET Median_Total_St ay = '"
.. $row3['Total_Stay'] . "' WHERE AdmitDoc = '" . $myDoc . "'";
mysql_query($Up Sql);
}//end if
$i++;
}//end while
}
else { //even number of admissions
$MyCounter = intval($No_Admi ts / 2);
while($row3=mys ql_fetch_assoc( $result2)){
if ($i==$MyCounter ){
$FirstTime = $row3['Total_Stay'];
}//end if
if ($i==$MyCounter + 1){
$SecondTime = $row3['Total_Stay'];
}//end if
$i++;
}//end while
$AvgVal = (strtotime($Fir stTime) + strtotime($Seco ndTime)) /
2;
$UpSql = "UPDATE tersummary SET Median_Total_St ay = '" .
date("H:i:s",$A vgVal) . "' WHERE AdmitDoc = '" . $myDoc . "'";
mysql_query($Up Sql);
}//end if/else
}
/*************en d block********** *************** ***/
Thanks,
Ross

There is no MEDIAN() operator in MySQL, so you will have to perform
this operation in several steps...

(1) Count the number of rows in each group
(2) For each group, if the count is odd, the median is the middle row
(when the set is ordered); when the count is even, the median is the
mean of the middle two rows.

In skeletal form...

// get counts for each set of interest...

SELECT AdmitDoc, COUNT(*) AS "Number_Adm its"
FROM tadmits
GROUP BY AdmitDoc

// for each AdmitDoc, work out LIMIT parameters (nb first row =
0)...

$doc = $datRow[ 'AdmitDoc' ];
$count = $datRow[ Number_Admits];

if $count is odd,
$limit1 = intval( $count / 2 );
$limit2 = 1;
else
$limit1 = intval( $count / 2 ) - 1;
$limit2 = 2;

// select middle row(s) from ORDERed set...

SELECT AdmitDoc,
SEC_TO_TIME(UNI X_TIMESTAMP(Out dt)-UNIX_TIMESTAMP( Decdt)) AS
"Doc_To_Adm it"
FROM tadmits
WHERE AdmitDoc = $doc
ORDER BY UNIX_TIMESTAMP( Outdt) - UNIX_TIMESTAMP( Decdt)
LIMIT $limit1, $limit2
if $count is odd,
// only one row...
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$median_doc_to_ admit = $datRow[ 'Doc_To_Admit' ];
else
// two rows
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val1 = $datRow[ 'Doc_To_Admit' ];
$datRow = mysql_fetch_arr ay( $datRows, MYSQL_ASSOC );
$val2 = $datRow[ 'Doc_To_Admit' ];
$median_doc_to_ admit = ( $val1 + $val2 ) / 2;

You'll have to do this separately for your two required measures
because they each need a different sort order.

---
Steve

[Steve]

Ross:

The outline I suggested uses PHP to control the loop through each
AdmitDoc, and to create the text of the query to find the median
value(s), so it is not a SQL-only solution. The main advantage it has
over your code is that using a LIMIT clause fetches only 1 or 2 rows
regardless of how many are in the table, which could be a sigificant
optimisation if you have thousands of rows.

Which bit wasn't working for you?

Here's a self-contained worked example...
<?php

$t_strServer = 'localhost';
$t_strUser = 'root';
$t_strPassword = '';
$t_objSQLConnec tion = mysql_connect( $t_strServer, $t_strUser,
$t_strPassword );
mysql_select_db ( 'TEST', $t_objSQLConnec tion );
mysql_query( 'DROP TABLE IF EXISTS tadmits', $t_objSQLConnec tion );
mysql_query( 'CREATE TABLE tadmits ( id INT, AdmitDoc VARCHAR(2), outdt
DATETIME, decdt DATETIME )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 1, "aa", "20041203000000 ",
"20041201000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 2, "aa", "20041204000000 ",
"20041202000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 3, "aa", "20041206000000 ",
"20041202000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 4, "bb", "20041208000000 ",
"20041203000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 5, "bb", "20041211000000 ",
"20041204000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 6, "cc", "20041210000000 ",
"20041206000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 7, "cc", "20041213000000 ",
"20041203000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 8, "cc", "20041216000000 ",
"20041205000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES ( 9, "cc", "20041223000000 ",
"20041201000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES (10, "cc", "20041222000000 ",
"20041207000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES (11, "dd", "20041226000000 ",
"20041202000000 " )', $t_objSQLConnec tion );
mysql_query( 'INSERT INTO tadmits VALUES (12, "ee", "20041214000000 ",
"20041203000000 " )', $t_objSQLConnec tion );

print "Doctor:\tMedia n days\n\n";

$t_datRows = mysql_query( 'SELECT AdmitDoc, COUNT(*) as "Number_Adm its"
FROM tadmits GROUP BY AdmitDoc', $t_objSQLConnec tion );
while( $t_datRow = mysql_fetch_arr ay( $t_datRows, MYSQL_ASSOC ) )
{
$t_strAdmitDoc = $t_datRow[ 'AdmitDoc' ];
$t_lngNumber_Ad mits = $t_datRow[ 'Number_Admits' ];
if( $t_lngNumber_Ad mits % 2 == 0 )
{
// even...
$t_lngLimit1 = intval( $t_lngNumber_Ad mits / 2 ) - 1;
$t_lngLimit2 = 2;
}
else
{
// odd...
$t_lngLimit1 = intval( $t_lngNumber_Ad mits / 2 );
$t_lngLimit2 = 1;
}

$t_datRows2 = mysql_query( 'SELECT AdmitDoc, UNIX_TIMESTAMP( outdt) -
UNIX_TIMESTAMP( decdt) AS "Doc_To_Adm it" FROM tadmits WHERE AdmitDoc =
"' . $t_strAdmitDoc . '" ORDER BY UNIX_TIMESTAMP( outdt) -
UNIX_TIMESTAMP( decdt) LIMIT ' . $t_lngLimit1 . ',' . $t_lngLimit2,
$t_objSQLConnec tion );

if( $t_lngNumber_Ad mits % 2 == 0 )
{
// even...
$t_datRow2 = mysql_fetch_arr ay( $t_datRows2, MYSQL_ASSOC );
$t_dblValue1 = $t_datRow2[ 'Doc_To_Admit' ];
$t_datRow2 = mysql_fetch_arr ay( $t_datRows2, MYSQL_ASSOC );
$t_dblValue2 = $t_datRow2[ 'Doc_To_Admit' ];
$t_lngMedianTim e = ( $t_dblValue1 + $t_dblValue2 ) / 2;
}
else
{
// odd...
$t_datRow2 = mysql_fetch_arr ay( $t_datRows2, MYSQL_ASSOC );
$t_lngMedianTim e = $t_datRow2[ 'Doc_To_Admit' ];
}

mysql_free_resu lt( $t_datRows2 );

// seconds to decimal days...
$t_lngMedianTim e = $t_lngMedianTim e / ( 3600 * 24 );

print "$t_strAdmitDoc :\t$t_lngMedian Time days\n\n";

}

mysql_free_resu lt( $t_datRows );

?>
I get this:

Doctor: Median days

aa: 2 days

bb: 6 days

cc: 11 days

dd: 24 days

ee: 11 days

---
Steve