Thank you. It makes sense now, the way you explain it :-)

 No doubt your variable is ordinal

According to your description your variable ordinal , there are no realy measurements to be scale .

Good Luck

thank you!!!

Ordinal

ordinal

If you end up with data coded on two levels, low and high, it is not really ordinal, but nominal, and you should treat it as such. 

When you transform the variable, do you consider the 0 original value as a missing value (as "not participating"), or not ("not participating" is meaningful) ?

If you end up with data on two levels, I agree with Dag.

Don't miss the fact that with SPSS, declaring a variable in a certain measurement level do restrict your choice to a predetermined set of statistical analyses.

So if you end up with three different values (or more), you can choose either to declare your recoded variable as an ordinal one, either as a nominal one, depending on the kind of statistical analysis you want to perform and the information you give (Is the order between categories, meaningful or not ?)

My original data is physiological measurements (arousal/reaction), numbers in millimeters that are measure  the highest spike. questions were three and the first highest was coded as 2, second as 1 and the other as 0 including reactions that were 0 (in excel). in spss the values for my data is 0 for for the lowest and 0 reaction, and 1 for the 1 and 2 reactions. some variables end up with two categories for deception detected (1) and deception not detected (0). but i also have one variable that includes the 0 as a category. the order is meaningful in my understanding. as for the tests i was thinking of performing t tests for comparing between variables (which represent the questions) and see the level of deception detected in each question in order to see if the difference is significant. however some people suggest mann whitney for this analysis but it is not clear to me yet.

your advice is truly appreciated

Determining the order of a group does not make sense in nominal variables (e.g. color, place of birth, etc.). According Your explanation, Your coded data are ordinal - You can determine the order of high and low reaction and it makes sense. 

But why You didn't use the original data - numbers in millimeters as continuous variable?

Hi Radovan, thanks for that.

the original data is deriving from 750 responses and each number is quite different, i coded it in spss to simplify things and to be able to set the values for deception detected and not detected (just two categories).

You can use the Mann-Whitney U statistic  to test the null hypothesis that two samples come from the same population (i.e. have the same median) or, alternatively, whether observations in one sample tend to be larger than observations in the other one.

For an introduction to the Mann-Whitney tests with SPSS, you can read the following technical note :

http://www.statstutor.ac.uk/resources/uploaded/mannwhitney.pdf

With regards to your last answer, I wonder if you could not have interest also in analyzing your original data before recoding (allowing you at least to justify the recoding thresholds, unless they do derive from theoretical considerations)

The variable is regarded as  ordinal variable . I prefer using the original (you said: data is deriving from 750 responses and each number is quite different) in order to get more accurate results by applying parametric methods after check your data.

Best Regards

I agree with two last comments: If you can use the original data in some form as predictor for the outcome false/true you could use the more powerful parametric tests. It would as well appear more objective than your classification, even if your scores are perfectly set.

thank you Steve,

a lot researchers of the field use logistic regression, that is true, but here i want to compare the efficiency in each repetition of the tests' questions. in addition there are no other independent variables.

Dear Calli,

I agree with Dag Sverre Syrdal ,

You data has 2 levels and its in the ordinal scale, and we need to apply non-parametric test such as Mann-Whitney's U test to compare the rankings among the groups.

Regards

Vishwanath