Time series with 3 factors from different sources

I've been asked to try to do a preliminary 'meta-analysis' of some data, but while comfortable with maths I don't have much experience with stats. I'm hoping to get some feedback with what I have attempted so far, to see if I'm on the right track.

Using SPSS for the software.

Design

One dependant variable 'size' taken from different studies performed over a range of years. Most studies did not go for more than one year.

Other factors:

• 'geography type' - all different types were sampled, some more than others.
• 'geography zone' - all of 4 different zones were sampled
• 'location' - many different locations were sampled, some more than others, some repeated over different years but for different studies, some only once!
• 'technique' - different sampling techniques used.

Questions

• Is size dependant on year?
• Does geography type, geography zone or technique affect size?
• How should data be combined? Should multiple measurements from the one location in one year be averaged?

Model

I think the model should be

• Dependant variable: size
• Fixed-effects: geography type, geography zone
• Random-effects: location, technique

Tests performed so far

Linear Regression:

1. Linear regression of size vs year showed significant trend. (0.002)
2. Linear regression of size vs year, when multiple measurements from the one location for one year were averaged (one data point per location), showed not-quite-significant trent (0.060)
3. Linear regression of size vs year, when averaged location values were averaged for the entire year (one data point per year), showed no significant trend

Question - which is more appropriate?

ANOVA:

1. Geography type vs size: Levene statistic reported significance of .000, so ANOVA with Games-Howell option checked was performed. Between groups sig of .000
2. Geography zone vs size: Levene statistic reported significance of .007, so ANOVA with Games-Howell option checked was performed. Between groups sig of .001
3. Technique vs size: Levene statistic reported significance of .000, so ANOVA with Games-Howell option checked was performed. Between groups sig of .001

Now I think some of these might be varying together.. How to check?

So what next?

EDIT: MADE A MISTAKE IN THIS SECTION, have now updated

Univariate GLM:

1. Fixed effects: geography type, geography zone; geo type sig (.014), zone not sig (.885), type * zone not sig (.239)
2. Fixed effects: year, geography type, geography zone; geo type sig (.002), zone not sig (.639), year sig (.000)... type * year also sig (.000)
3. Fixed effects: year, geography type, geography zone; random effects, technique; SPSS could not calculate error term.
4. Fixed effects: year, geography type, geography zone, technique; type, year and technique all significant. type * year also significant.

Is this the correct approach? Is it possible that there is not enough data?

When comparing means just looking at zone, there seems to be a difference. With the GLM it seems zone doesn't matter? How is this possible?

-