Scoring survey questionnaire responses

[Cruise]

Anyone could help understand and make sense of a standard code below?

I have a survey data that has 8 items (series of questions). I chose one of them (PF=physical functionality) for demo purpose and attached for sample data. It's consisted of 10 questions (PF01-PF10) and has 3 response options.

      0-No, not limited at all

      1-Yes, limited a little

      2-Yes, limited a lot

I'd like to score the responses and below is the standard code recommended. 

ARRAY PFI(10) PF01-PF10;
PFNUM = N(OF PF01-PF10);
IF PFNUM GE 5 THEN RAWPF = SUM(OF PF01-PF10);
PF = ((RAWPF - 10)/(30-10)) * 100;
LABEL  PF = 'PHYSICAL FUNCTIONING (0-100)'
       RAWPF = 'RAW PHYSICAL FUNCTIONING';

i'm fine with this part to treat the subject's response for the item missing unless responded to mor ethan 5 out of 10 questions.

IF PFNUM GE 5 THEN

However, the code part below doesn't make sense to me.

PF = ((RAWPF - 10)/(30-10)) * 100;

I keep asking myself:

Overall, how part: PF = ((RAWPF - 10)/(30-10)) * 100; make sense, conceptually?

Why would you subtract levels of series of question (10) from the total amount of response values (RAWPF) ?

Why would you subtract levels of series of question (10) from the total number of possible responses (10*3=30)?

Anyone experienced in scoring survey data? and let me know if this approach makes sense to you? if yes, do you mind explaining how? if not, any better approach you'd recommend? I did a lit review with no success yet.

Once all responses are scored I will Z-transform them so that standardized response scores can be summed across different items.

[Jagadishkatam · Accepted solution]

The PF-10 is one of the questions in SF-36 questionnaire, which includes the following questions about activities you might do during a typical day.  They have the following values

     3-No, not limited at all

      2-Yes, limited a little

      1-Yes, limited a lot

ACTIVITIES	Yes, Limited A Lot	Yes Limited A Little	No, Not Limited At All
a.       Vigorous activities, such as running,               lifting heavy objects, participating in          strenuous sports	1	2	3
b.      Moderate activities, such as moving a       table, pushing a vacuum cleaner,       bowling, or playing golf	1	2	3
c.    Lifting or carrying groceries	1	2	3
d.    Climbing several flights of stairs	1	2	3
e.    Climbing one flight of stairs	1	2	3
f.    Bending, kneeling or stooping	1	2	3
g.    Walking more than a mile	1	2	3
h.    Walking half a mile	1	2	3
i.     Walking one hundred yards	1	2	3
j.     Bathing or dressing yourself	1	2	3

Scale Scoring for PF

Step 1- Sum up final item scores

       Lowest possible raw score = 10; Highest possible raw score = 30

       Possible raw score range = 20

Step 2- Following standard formula is used to calculate raw scale scores

Transeformed scale = [(Actual raw score- lowest possible raw score)/Possible raw score range x 100

e.g.1- PF of someone with the lowest possible raw score = [(10-10)/20] x 100 = 0

e.g.2- PF of someone with the highest possible raw score = [(30-10)/20] x 100= 100

PF = ((RAWPF - 10)/(30-10)) * 100;

So here the RAWPF=Actual raw score, 10=lowest possible raw score, 30-10=Possible raw score range which is 20.

Please find attached E2 document for more details

[Cruise]

@Jagadishkatam

Hi,

Document attached is directly relevant to my data which is veteran's SF-36. Do you have continuation of the document or any other resources recommendable that explains methods to help summarize Physical and Mental component scales?

[Jagadishkatam]

@Cruise I checked for more details however I could find only this document. However you could check the link below

https://en.wikipedia.org/wiki/SF-36#Scoring

and further links in this site.

[Reeza]

Given @Jagadishkatam response I'm not sure your approach of standardization to Z scores is correct. I would review the methdology used in papers for this data where they likley explain it, but I would guess the PF scores are essentially standardized, but via a different methodolgy. 

[Cruise]

@Reeza

I am so glad you brought this out. According to the attached guideline, means and STDEV used in Z-transformation formulas are derived from MOS-SF36 means and STDEV from US population. However, my data is not MOS-SF36 but veteran's version VR-SF36. They're slightly different from each other in terms of number of multiple choices on two question items. I don't want to convert my data to MOS-SF36. I also am not concerned too much of external validity. Because it's a pilot clinical trial of small sample size (n=25) where internal comparisons (pre and post test) is more of importance.

PF_Z=(PF-84.52404)/22.89490;
rp_z=(rp2-81.19907)/33.79729...;

  praw=(PF_Z * .42402)+(RP_Z * .35119)...;
  mraw=(PF_Z * -.22999)+(RP_Z * -.12329)+...;

  PCS = (praw*10) + 50;
  MCS = (mraw*10) + 50;

  label PCS='STANDARDIZED PHYSICAL COMPONENT SCALE'
        MCS='STANDARDIZED MENTAL COMPONENT SCALE';

Any alternative approach that would allow summarizing different items to PCS and MCS Component Scales (as in above code)?