You need check if

1.       Is there difference between the side was in a diagonal and others

Null hypothesis: they are the same

Alternative: they are different

Assume the one in diagonal as true value, using t-test

If not the same, shows big different, you should consider others only.

2.       When you using a linear regression,

The variable used in the left is outcome or dependent variable

The variables used in the left of the = are predictors of independent variables

Hypothesis:

The effect of independent variable is zero (slope or regression coefficient  =0),

You can think a few of them, which are interested by you, are predictors, the others are control factors

3.       Those factors on the left side, you may do one by one:

If numerical with many different values, you could use linear regression, but you need check the residual =observation-predicted values from model are normal distributed, mean =0 and the variance are constant.  Random distributed around zeroes

If not, transformation is needed.

Doing so, the total independent factors for your data should be no more than 9: when you count

·         Interception is one

·         Numerical variable  is one

·         If a categorical factor has m categories, count m-1.

4.       If the one used in left side: are counts like integers,

Poisson regression can be used.

If all predictor factors are categorical data, log-linear model is also ok

5.       If the outcome is binary: buried or not

Logistic regression

6.       If time to buried is also considered: survival analysis: Kaplan-Meir or Proportional Hazard modeling

Hope it help.

How far apart are your sites? 

Did you use multiple linear regression or a series of simple linear regressions? 

What is(are) your response variable(s)? Is it number of cockles? If so, that uses a poisson regression. 

The model selection depends on test hypothesis and data structure. Please state

what do you want to test?

what is your outcome: number of cockles by adult and juvenile?

the types of the predict factors: numerical or category, if category, how many levels?

Without the above information, it is difficult to have suitable suggestion.

As my previous speakers said, please give a bit more information on what your outcome variable is and what your independent variables are. Whats the research question you want to address?

Hi there thanks so much!! ok 

Site locations are approximaetly 2-3m apart, one side was in a diagonal across the mudflats while the other was done horizntal to the shore line, i'd say an area of approx. 20m x 50m maybe more? but still have to work out exactly.

A simple linear regression was used.

There is no test hypothesis as yet, this needs to be defined, which is what I am struggling with, and how to test and analyses the data. (part of the brief, so i am trying to look at all angles to pull hypothesis,data analysis etc together)

the outcome is up to us, we have both numbers (adult and juvenile) as well as an ambient number of cockles on the second data collection set.

What do you mean by predict factors? sorry the data i have given is what we have to work with. Do we must define and design our study from the information i have provided, does that help?

 What i have noticed from simply looking at the data is;

That water coverage seems to correlate with burial rates. (we timed how quickly they buried once marked and relocated in trial plots)

We got a activity rate by adding how many cockles buried plus how many marked cockles were still in our trial plots area after 24 hours. We also then counted the total cockle population in the trial plots to give us ambient numbers.

I just need some suggestions on how to I may be able to analysis this data but I am struggling with a hypothesis, etc so I am looking for directions which may help me define all the above. many thanks to you all!! cheers Kiri

You need check if

1.       Is there difference between the side was in a diagonal and others

Null hypothesis: they are the same

Alternative: they are different

Assume the one in diagonal as true value, using t-test

If not the same, shows big different, you should consider others only.

2.       When you using a linear regression,

The variable used in the left is outcome or dependent variable

The variables used in the left of the = are predictors of independent variables

Hypothesis:

The effect of independent variable is zero (slope or regression coefficient  =0),

You can think a few of them, which are interested by you, are predictors, the others are control factors

3.       Those factors on the left side, you may do one by one:

If numerical with many different values, you could use linear regression, but you need check the residual =observation-predicted values from model are normal distributed, mean =0 and the variance are constant.  Random distributed around zeroes

If not, transformation is needed.

Doing so, the total independent factors for your data should be no more than 9: when you count

·         Interception is one

·         Numerical variable  is one

·         If a categorical factor has m categories, count m-1.

4.       If the one used in left side: are counts like integers,

Poisson regression can be used.

If all predictor factors are categorical data, log-linear model is also ok

5.       If the outcome is binary: buried or not

Logistic regression

6.       If time to buried is also considered: survival analysis: Kaplan-Meir or Proportional Hazard modeling

Hope it help.

Are all of your sites on the same shoreline? For the analysis part, Yuan has you there. From the design standpoint, If all of your sites are on the same shoreline, I would include a categorical variable, Site ID. If you find out they were 9 sites at 5 different "beaches", then you will use a split plot design.

The data you are talking about sounds eerily familiar to me.  I think authors with similar data used a cluster analysis method on their data. 

yes same shoreline, Whangateatu estuary Leigh, New Zealand, data is part of our estuarine ecology paper and was done last month.

I have done a simple linear regression which has given me a negative relationship with ambient cockles and marked cockles. Activity rate & water coverage has given me a positive relationship, the activity rate and compactability ( a course weight dropped into substrate and measured)  has also given me a positive but weak relationship. 

my hypothesis is based around ' ambient density and the influence it has on cockle movement' just rough at this stage so i am still trying to refine the hypothesis and how i look at analysis from here in. many thanks for all your help, other members of my class may have posted similar questions here? 

will look at the above mentioned  suggestions and see how these correlate/refine my work. many thanks Kiri

Kiri,

Form what you did on linear regression is the same as Pearson Correlation Coefficient(PCC), among paris of variables. You can trasfer the slope in simple linear regression to Pearson Correlation Coefficient. Y=intercept+X*slope, then slope=PCCof X&Y*Sy/Sx.

Sy and Sx are standard deviation of Y and X respectively.

In fact, you can get PCC by use any Stat Package, even by hand calculation.

If PCC near 1 or -1, say >.8 or

Ok great, 

I did a PCC on burial rate and horizontal movement ( which i added together to get a activity rate) and the co efficient came out as 0.132 between the two types of movements, what is this telling me? 

I then  did  a PCC on the activity rate and compactibility mm and got 0.452. I did this in excel. Therefore could I say, if I transfer compactability to muddiness of sediment, that movement and sediment type are highly correlated from what you have written above?

  When you write about multiple linear regression and not putting both of them as independent factors, can you clarify for me?

I have been using activity rate as the intercept and compactibility, shell hash, water coverage as  slope, is this correct? Shell hash and water coverage were assessed as high, low medium for shell hash and partial, full to none for water coverage. I converted these  to 1, 2, 3 for shell hash estimates and 0, 50% and 100% for water coverage to be able to analyse in linear regression in excel, do you think this is appropriate?

I also did simple linear regression on marked cockles compared to ambient cockle numbers to look at ambient population dynamics/influences in the corer. We used a 13cm corer at each  original test plot then used a 57cm sieve as our external perimeter, to see how many cockles had moved out of our original corer plot. The corer data was where I got my negative relationship between our marked and ambient cockles. However I couldn't do it for the larger perimeter as the ambient numbers were to big, so the marked cockles weren't showing up on any scatter plots.

I have found that the average density for our site is .74/m sq.  We recovered 69% of cockles in our corer plots after 24hrs, and in the sieve plots 11.5% recovered with 19.5%  unaccounted for or not found - they could be outside the plot area or our nail polish was washed off or fell off. so our average recovery rate for marked cockles was 80.44% 

thanks so much for all your help it is really great to have the feedback and guidance!! thank you

Kiri 

Kiri,

See my answers follow your statement:

I did a PCC on burial rate and horizontal movement ( which i added together to get a activity rate) and the co efficient came out as 0.132 between the two types of movements, what is this telling me?

Yuanzhang you are amazing!!! thankyou so much, I will try to get that software and follow the leads you have given me!! thank you so much!!!!  I will press on and let you know how it has worked out, probably ask more questions!!! thankyou so much!!!