Error with I-squared in metan command

[Sharon Phillips]

I have been getting a 0 for i-squared for multiple analyses when I know this is not the correct value. Does anyone know of any bugs that could cause this or a way I've set up my data that could cause it? My code is below

Stata V13
metan loghr loghr_lower loghr_upper if comparison ==1, eform random effect(AdjHR) lcols(Study) title (DMPA) xlabel (0.2,1,2,5) force textsize(150)

Thanks!

[Sharon Phillips]

Sorry, here is the full output
. metan loghr loghr_lower loghr_upper if comparison ==5, eform random effect(AdjHR) lcols(Study Year) favours (Favours benefit # Favours harm) title (OCPs) xlabel (0.5,1,5) force textsize(150)

Study | ES [95% Conf. Interval] % Weight
---------------------+---------------------------------------------------
Wall | 1.390 0.900 2.150 18.12
Balkus | 0.760 0.370 1.550 6.70
Morrison | 0.880 0.490 1.300 14.44
Heffron | 1.800 0.550 5.820 2.47
Reid | 0.910 0.450 1.830 6.98
Morrison | 0.990 0.690 1.420 26.38
Baeten | 1.460 1.000 2.130 24.04
Myer | 0.660 0.090 4.780 0.87
---------------------+---------------------------------------------------
D+L pooled ES | 1.122 0.932 1.351 100.00
---------------------+---------------------------------------------------
Heterogeneity calculated by formula
Q = SIGMA_i{ (1/variance_i)*(effect_i - effect_pooled)^2 }
where variance_i = ((upper limit - lower limit)/(2*z))^2

Heterogeneity chi-squared = 6.58 (d.f. = 7) p = 0.474
I-squared (variation in ES attributable to heterogeneity) = 0.0%
Estimate of between-study variance Tau-squared = 0.0000

Test of ES=1 : z= 1.22 p = 0.222

.
end of do-file

[Nick Cox]

Could you please copy and paste output into the forum software within CODE delimiters? I tried copy and paste from your post, but too many spaces had disappeared into oblivion for that to work.

See also http://www.statalist.org/forums/help#stata

[Sharon Phillips]

Thank you for the correction. I am using Stata 13.1 for this meta-analysis. LogHR is log hazard ratio, loghr_lower is lower bound, loghr_upper is upper bound, comparison is the outcome of interest.

metan loghr loghr_lower loghr_upper if comparison ==1, eform randomi effect(AdjHR) lcols(study) title(DMPA) xlabel (0.2,1,2,5) force textsize(150)

           Study     |     ES    [95% Conf. Interval]     % Weight
---------------------+---------------------------------------------------
Morrison             |  1.690       1.020     2.780          6.58
Wall                 |  1.340       0.850     2.120          7.92
Crook                |  1.450       1.090     1.930         20.25
McCoy                |  1.280       0.900     1.820         13.33
Morrison             |  1.280       0.920     1.780         15.18
Heffron              |  2.040       0.810     5.150          1.93
Morrison             |  1.250       0.890     1.780         13.76
Baeten               |  1.730       1.280     2.340         18.17
Kleinschmidt         |  0.460       0.060     3.790          0.38
Myer                 |  0.750       0.330     1.680          2.50
---------------------+---------------------------------------------------
D+L pooled ES        |  1.401       1.232     1.593        100.00
---------------------+---------------------------------------------------
 Heterogeneity calculated by formula
  Q = SIGMA_i{ (1/variance_i)*(effect_i - effect_pooled)^2 } 
where variance_i = ((upper limit - lower limit)/(2*z))^2 

  Heterogeneity chi-squared =   7.47 (d.f. = 9) p = 0.588
  I-squared (variation in ES attributable to heterogeneity) =   0.0%
  Estimate of between-study variance Tau-squared =  0.0000

  Test of ES=1 : z=   5.14 p = 0.000

[David Fisher]

Dear Sharon,
I have copied-and-pasted your study results into my Data Editor (14.0) and have re-run the meta-analysis using three different packages. I-squared is definitely zero with the data as presented.
When you say "I know this is not the correct value", what is your source for that? Have you previously analysed the same data and got a different result? If so, are you absolutely sure the analyses and data were identical on both occasions? (e.g. if/in restrictions, log transformations, adjusted vs unadjusted, different random-effects method)
Thanks,
David.