Hello, Can someone kindly explain what the best next step is, please?
I measured arsenic concentrations within solutions after application of different treatment at different time intervals.
It has been suggested to me that I use a two-way ANOVA to analyse my results. I ran a GLM with time and treatment as factors against As concentration.
My results showed
Factors ----------df----------As
Treatment-----------4------------10.34***
Time -------------------5------------32.41***
Treatment*Time-20-----------1.01*
*(p
Use simple effects analysis. To this, you need to add a line like the following into your syntax, since SPSS does not have a click and do option for this
/EMMEANS=TABLES(Treatment*Time) Compare(Time) ADJ(SIDAK)
or
/EMMEANS=TABLES(Treatment*Time) Compare(Treatment) ADJ(SIDAK)
PS: SIDAK is my choice, there are other alternatives.
Yes, you can multiple comparison methods as(Duncan or LSD,sheffe) to determine which level of factore is more effect.
With Best Regrads
In my opinion, the interaction itself matters. in such a design (treatments and times crossed) interaction means that the evolution with time is not the same for different treatments, that sounds like the point of interest.
If you are interested only in the final results of the treatments, the intermediate values are superfluous, just analyse the last one.
So, we have a two-factorial design (factors "treatments" and "time"). With 4 ”treatments" and 5"time” this is a 4x5 design and 20 possible combination of the factor levels.
You can create new variable represent the combination treatments ( interactions) in a new column e.g a1b1, a1b2...... and then you can comparison this 20 groups with Tukey's HSD (if the number of comparisons is more than 5 and you want to check all possible pairwise combinations), Bonferroni or other post hoc.
Best regards
Depending on the questions you are looking to answer, the number of replications (if you have more than one), and your experience on the subject you are working, you could first select what significance level is important for you and use the same across main effects and interaccions (i.e. relevant for your experiments). Thus, at a 0.01 level, the interaction would be not significant, and you could proceed to evaluate the main effects independently, and use the mean comparisons methods (Duncan, Tukey, Shchaffe) straightforward for each factor. A useful think is to make a graph of a factor against the other and examine the possible interactions.
If interactions are significant, analysis is more complex.
On the other hand, although I cannot infer from your data the exact layout of the experiment, the factor time could be a problem if you are changing the concentration of solution in the same recipient (replication or subject). In thiis case, you have a repeated measures problem, and autocorrelation in time could occur. Overall, you could do a residual analysis to ensure you are meeting the factorial model assumptions i.e. residuals distribute normal with mean 0, common variance, and independent residuals.
Regards.
Um... Looks like a completed analysis to me. I assume that the numbers under As are the coeffients from the model. Clearly Time and Treatment have a very significant affect on concentration. By adding the interaction term you can judge just how much time and treatment interact with one another to affect the concentration - the 1.01 suggests that the interaction is not profound but the marginally significant p-value says to keep it in the analysis. Doing this accounts for the interaction and gives you coefficients for Time (32.41) and Treatment (10.31) that are now independent of the interaction. You have controlled for a small but significant interaction which means the the Time and Treatment effects are more accurate than if you had left the interaction term out. This model is clearly the best way to determine the effects. You might get slightly different coefficients for time and treatment if you did a simple regression analysis but they would not be controlled for interaction and so would not be as accurate as the one from the full model you have tried here.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS