questionnaire

[Anata Ionescu]

I`m in 2nd year (Bologna Bachelor`s Level) at Psychologies and I`ve been assigned to make a questionnaire. It is intended to measure self-disclosure: 1 means low self-disclosure, 5 means high self-disclosure and yes, it is a summative scale. My teacher said that items need to be normally distributed, and he said it clearly, I don`t really intend to argue about that.
What do you think?

Thank you.

[Art Kendall]

In most instances, the concern is that residuals not be severely discrepant from normality.  If you wish to do as your teacher asks with regard to raw item score normality, you might apply a similar approach, retain items that are not severely discrepant from normally distributed. When you run the RELIABILITY procedure, look to see that Standard deviations are not very small.  With a 1 to 5 response scale  the most extreme situation possible (unless there is no variance, in which situation the item is useless) is shown via the syntax example below. That is why item normality is not something to worry about.
open a new instance of SPSS. copy and paste the syntax into a syntax window. Run it.  Note that the standard errors of skewness and kurtosis are nowhere near 1.96. It would be unusual for such an item to increase the alpha for the scale.

data list list/ item (f1) kount(f2).
begin data
1 32
5 1
end data.
weight by kount.
descriptive vars= item /statistics= all.

Art Kendall
Social Research Consultants

Anata Ionescu wrote:

I`m in 2nd year (Bologna Bachelor`s Level) at Psychologies and I`ve been assigned to make a questionnaire. It is intended to measure self-disclosure: 1 means low self-disclosure, 5 means high self-disclosure and yes, it is a summative scale. My teacher said that items need to be normally distributed, and he said it clearly, I don`t really intend to argue about that.
What do you think?

Thank you.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Art Kendall]

clarification.

That is why item normality is not something to worry about.

That is why item normality is not something to worry about beyond the rule of thumb you posted in your original.  You would have removed the item simply by looking at the distribution.

Note that the standard errors of skewness and kurtosis are nowhere near 1.96.

I misread the output. I was thinking of some old procedures that output skewness/se.

if you use FREQUENCIES variables = ... /statistics = all
  /HISTOGRAM NORMAL.
you should have the info you need for a first screening of items.
As an exercise create 2 scales when you do RELIABILITY.  One with all 39 items and one with the normality screen applied. pay attention to the squared multiple correlation, and the alpha-if-item-deleted.

Art Kendall

Art Kendall wrote:

In most instances, the concern is that residuals not be severely discrepant from normality.  If you wish to do as your teacher asks with regard to raw item score normality, you might apply a similar approach, retain items that are not severely discrepant from normally distributed. When you run the RELIABILITY procedure, look to see that Standard deviations are not very small.  With a 1 to 5 response scale  the most extreme situation possible (unless there is no variance, in which situation the item is useless) is shown via the syntax example below. That is why item normality is not something to worry about.
open a new instance of SPSS. copy and paste the syntax into a syntax window. Run it.  Note that the standard errors of skewness and kurtosis are nowhere near 1.96. It would be unusual for such an item to increase the alpha for the scale.

data list list/ item (f1) kount(f2).
begin data
1 32
5 1
end data.
weight by kount.
descriptive vars= item /statistics= all.

Art Kendall
Social Research Consultants

Anata Ionescu wrote:

I`m in 2nd year (Bologna Bachelor`s Level) at Psychologies and I`ve been assigned to make a questionnaire. It is intended to measure self-disclosure: 1 means low self-disclosure, 5 means high self-disclosure and yes, it is a summative scale. My teacher said that items need to be normally distributed, and he said it clearly, I don`t really intend to argue about that.
What do you think?

Thank you.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[John F Hall]

...and don't forget that the more items you add together the closer the resultant score will approximate to a normal distribution, regardless of whether the items measure the same thing or not.

----- Original Message -----

From: [hidden email]

To: [hidden email]

Sent: Monday, December 07, 2009 2:16 PM

Subject: Re: questionnaire

clarification.

That is why item normality is not something to worry about.

That is why item normality is not something to worry about beyond the rule of thumb you posted in your original.  You would have removed the item simply by looking at the distribution.

Note that the standard errors of skewness and kurtosis are nowhere near 1.96.

I misread the output. I was thinking of some old procedures that output skewness/se.

if you use FREQUENCIES variables = ... /statistics = all
  /HISTOGRAM NORMAL.
you should have the info you need for a first screening of items.
As an exercise create 2 scales when you do RELIABILITY.  One with all 39 items and one with the normality screen applied. pay attention to the squared multiple correlation, and the alpha-if-item-deleted.

Art Kendall

Art Kendall wrote:

In most instances, the concern is that residuals not be severely discrepant from normality.  If you wish to do as your teacher asks with regard to raw item score normality, you might apply a similar approach, retain items that are not severely discrepant from normally distributed. When you run the RELIABILITY procedure, look to see that Standard deviations are not very small.  With a 1 to 5 response scale  the most extreme situation possible (unless there is no variance, in which situation the item is useless) is shown via the syntax example below. That is why item normality is not something to worry about.
open a new instance of SPSS. copy and paste the syntax into a syntax window. Run it.  Note that the standard errors of skewness and kurtosis are nowhere near 1.96. It would be unusual for such an item to increase the alpha for the scale.

data list list/ item (f1) kount(f2).
begin data
1 32
5 1
end data.
weight by kount.
descriptive vars= item /statistics= all.

Art Kendall
Social Research Consultants

Anata Ionescu wrote:

I`m in 2nd year (Bologna Bachelor`s Level) at Psychologies and I`ve been assigned to make a questionnaire. It is intended to measure self-disclosure: 1 means low self-disclosure, 5 means high self-disclosure and yes, it is a summative scale. My teacher said that items need to be normally distributed, and he said it clearly, I don`t really intend to argue about that.
What do you think?

Thank you.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD