ROC

[Chelminski, Iwona]

Hi group,
I have two instruments-one (let's call it X) measures a number of different
symptoms of an illness and the other measure (let's call it Y) allows to
rate an overall impression of the illness (there are 7 numbers on the Y that
capture 4 levels of the illness: none (1-2), mild (3-4), moderate (5),
severe (6-7). I want to identify the cut off scores on the X total score for
adjacent severity levels on the Y by using ROC analyses.
Should I include only these cases who scored within those adjacent levels,
or all cases? For example, for None vs. Mild I'm including only those that
scored 1 or 2 (=none) or 3 or 4 (=mild) on the Y (i.e, excluding those who
scored 5, 6, and 7), right?
Or should I use all cases and split the sample between "1 or 2" (for none)
and "GE 3" (for the rest)?
Thanks in advance!!!
Iwona

[Chelminski, Iwona]

Hi Folks,
I'm resending this, hoping that someone might give me a suggestion (please
:-)
Thanks in advance

>  -----Original Message-----
> From:         Chelminski, Iwona
> Sent: Thursday, September 14, 2006 1:35 PM
> To:   SPSS list (E-mail)
> Subject:      ROC
>
> Hi group,
> I have two instruments-one (let's call it X) measures a number of
> different symptoms of an illness and the other measure (let's call it Y)
> allows to rate an overall impression of the illness (there are 7 numbers
> on the Y that capture 4 levels of the illness: none (1-2), mild (3-4),
> moderate (5), severe (6-7). I want to identify the cut off scores on the X
> total score for adjacent severity levels on the Y by using ROC analyses.
> Should I include only these cases who scored within those adjacent levels,
> or all cases? For example, for None vs. Mild I'm including only those that
> scored 1 or 2 (=none) or 3 or 4 (=mild) on the Y (i.e, excluding those who
> scored 5, 6, and 7), right?
> Or should I use all cases and split the sample between "1 or 2" (for none)
> and "GE 3" (for the rest)?
> Thanks in advance!!!
> Iwona
>

[Spousta Jan]

Hi Iwona,

In my opinion, the best way is to keep all cases and make 3 subsequent
analyses:

* 1 or 2 against 3+
* 1 - 4 against 5+
* 1 - 5 against 6+

Otherwise you'll throw out information without reason.

Greetings

Jan


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Chelminski, Iwona
Sent: Friday, September 15, 2006 5:02 PM
To: [hidden email]
Subject: Re: ROC

Hi Folks,
I'm resending this, hoping that someone might give me a suggestion
(please
:-)
Thanks in advance

>  -----Original Message-----
> From:         Chelminski, Iwona
> Sent: Thursday, September 14, 2006 1:35 PM
> To:   SPSS list (E-mail)
> Subject:      ROC
>
> Hi group,
> I have two instruments-one (let's call it X) measures a number of
> different symptoms of an illness and the other measure (let's call it
> Y) allows to rate an overall impression of the illness (there are 7
> numbers on the Y that capture 4 levels of the illness: none (1-2),
> mild (3-4), moderate (5), severe (6-7). I want to identify the cut off

> scores on the X total score for adjacent severity levels on the Y by
using ROC analyses.
> Should I include only these cases who scored within those adjacent
> levels, or all cases? For example, for None vs. Mild I'm including
> only those that scored 1 or 2 (=none) or 3 or 4 (=mild) on the Y (i.e,

> excluding those who scored 5, 6, and 7), right?
> Or should I use all cases and split the sample between "1 or 2" (for
> none) and "GE 3" (for the rest)?
> Thanks in advance!!!
> Iwona
>