Which test? Values differ from zero significantly?

[Ibrahim Filiz]

Dear all,

how I can check that the following values differ from zero significantly?

0 = expected value
It is one-sample

60
10
0
-10
-20
-10
20
17
-30
-246
-10
-31
0
-10
10
-34
30
-6
0
0
-5
30
-10
-10
10
-40
-10
-20
10
-6
88
86
-2
11
-25
-40
0
-10
25
-10
-81
30
0
40
4
-10
14
0
40
40
49
20
19
-69
41
-86
-10
20
-20
0
-10
-10
-10
30
0
0
0
0
30
-40
-10
25
40
70
-10
-30
40
10
-20
0
44
-20
20
-57
40
-50
-40
-10
10
0
10
30
30
14
0
-192
30
0
10
-10
20
-40
0
-251
0
0
29
-20
-246
14
0
10
-10
-62
0
0
10
-10
0
29
10
0
-30
-10
0
-10
40
-216
-40
20
-10
0
-10
40
10
-246
10
0
0
-213
0
-10
0
20
-10
0
-40
-10
0
40
0
-10
0
-10
10
-246
0
-246
-6
78
20
20
30
88
2
-10
0
20
0
20
40
-10
0
10
10
77
10
-36
20
30

Thnaks,
Ibrahim

[Carlo Lazzaro]

Ibrahim.
assuming that your variable is named -var-:

input var
60
10
0
-10
-20
-10
20
17
-30
-246
-10
-31
0
-10
10
-34
30
-6
0
0
-5
30
-10
-10
10
-40
-10
-20
10
-6
88
86
-2
11
-25
-40
0
-10
25
-10
-81
30
0
40
4
-10
14
0
40
40
49
20
19
-69
41
-86
-10
20
-20
0
-10
-10
-10
30
0
0
0
0
30
-40
-10
25
40
70
-10
-30
40
10
-20
0
44
-20
20
-57
40
-50
-40
-10
10
0
10
30
30
14
0
-192
30
0
10
-10
20
-40
0
-251
0
0
29
-20
-246
14
0
10
-10
-62
0
0
10
-10
0
29
10
0
-30
-10
0
-10
40
-216
-40
20
-10
0
-10
40
10
-246
10
0
0
-213
0
-10
0
20
-10
0
-40
-10
0
40
0
-10
0
-10
10
-246
0
-246
-6
78
20
20
30
88
2
-10
0
20
0
20
40
-10
0
10
10
77
10
-36
20
30
end
ttest var == 0

. ttest var == 0

One-sample t test
------------------------------------------------------------------------------
Variable |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
---------+--------------------------------------------------------------------
     var |     180   -8.133333      4.3855    58.83766   -16.78726    .5205983
------------------------------------------------------------------------------
    mean = mean(var)                                              t =  -1.8546
Ho: mean = 0                                     degrees of freedom =      179

    Ha: mean < 0                 Ha: mean != 0                 Ha: mean > 0
 Pr(T < t) = 0.0326         Pr(|T| > |t|) = 0.0653          Pr(T > t) = 0.9674

[Ibrahim Filiz]

Dear Carlo,

I did the same test, but I was thinking that I can´t work with mean comparison tests. Should I calculate the absolute values and then do an different test?

Another question, I have 8 different rows of the same sample. I asked 8 times the same subjects ragarding forecasts. Should I do 8 test or is that possible to merge the rows?

Thanks,
Ibrahim

[Carlo Lazzaro]

Ibrahim:
if you have repeated measures about the same item on the same sample, you have, in all likelihood, a panel data structure (see -xtreg-).
However, you do not tell us nothing about regression predictors.
Please provide further details. Thanks.

[Nick Cox]

That's a very odd distribution.

quantile var, rlopts(lc(none)) ms(Oh) yla(-250(50)100, ang(h))

[Ibrahim Filiz]

Dear Carlo and Nick,

I attached my data. Short description: The subjects should forecast stock prices for the next 10 periods. They knew the expected value in each round. The rational strategy equals the expected value.

In the rows, I subtracted the forecasts with the expected values in each round. You see the difference in the rows. Now, I would like check, are the subjects following the rational strategy? Differ the values from zero...

Thanks,
Ibrahim

[Nick Cox]

Thanks for the detail, but I won't (indeed at the moment can't) look at a spreadsheet file. We do explain this (FAQ Advice #12).

The more important point is that several low values < -100 are heavily influential in the test.

[Ibrahim Filiz]

I would like test the sample in all - to test each period is maybe not the best solution. You think, categorical test is fine? Or any other idea?

Thanks,
Ibrahim

[Joro Kolev]

Seems that you are beyond the single sample situation, and you need some sort of regression.

Read -dataex- and post some representative sample of your data, if you have 8 varialbes, show how those 8 look like and what you want to test.

If you are worried for robustness of the procedure, you can use robust regression or median regression.