I give below the changes in the codes that I made for questions 2-4. Of course, there are other ways to do these tings, and so my codes are just suggestions.

@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
Question 2: Alternative Specific Constants

ADDED TO logit.m:


%Do you want to include alternative specific constants for all except the last alt?
%ASC=1 for yes, and ASC=0 for no

ASC=1;


ADDED to doit.m:

%Create ascs
if ASC == 1;
  nalt=max(XMAT(:,2));
  startasc=zeros(1,nalt-1);
  for n=1:(nalt-1);
    newvar=XMAT(:,2)==n;
    VARS=[VARS newvar];
    NAMES=[NAMES num2str(n)];
  end;
end;

if ASC==0;
   param=[B]';  %starting values: take transpose since must be col vector
else;
   param=[B startasc]';
end;

@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
Question 3: Scale

ADDED TO logit.m:

global WANTSC SCGROUP

%Do you want to estimate the scale for a group?
% WANTSC=1 for yes, WANTSC=0 for no
WANTSC=1;

%create a variable that identifies the group with an estimated scale.
%SCGROUP must have as NROWS rows and 1 col, with elements of 1 for in group, 0 not
SCGROUP=XMAT(:,11)+XMAT(:,12); 



ADDED TO doit.m:

if WANTSC==1;
   B=[B ones(1,1)];
   NAMES=[NAMES 'scale'];
end;

ADDED TO loglik.m:

global WANTSC SCGROUP

AND replaced "v=VARS*coef" with this:

if WANTSC==0;
  v=VARS*coef;
else;
  v=VARS*coef(1:end-1,1);
  v(SCGROUP==1,1)=v(SCGROUP==1,1).*coef(end,1); %Multiplicative scale for group
end;

@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
Question 4: Gradient

CHANGED "options" and fminunc" commands in doit.m to:

options=optimset('LargeScale','off','Display','iter','GradObj','on',...
    'MaxFunEvals',10000,'MaxIter',MAXITERS,'TolX',PARAMTOL,'TolFun',LLTOL,'DerivativeCheck','on');
[paramhat,fval,exitflag,output,grad,hessian]=fminunc(@loglikgr,param,options);

HERE is the entire loglikgr.m code:

%Inputs: coef is Kx1 where K is number of explanatory variables
%
%Globals: NCS: scale number of choice situations
%         IDCASE: NROWSx1 vector identifying the rows for each choice situation (1-NCS), where NROWS is number of alternatives
%                          in all choice situations combined
%         IDDEP: NROWSx1 vector identifying the chosen alternative (1=chosen, 0=nonchosen)
%         VARS: NROWSxK matrix of explanatory variables

function [ll,gr]=loglikgr(coef);

global NCS IDCASE IDDEP VARS

p=zeros(NCS,1);
k=size(coef,1);
g=zeros(NCS,k);
v=VARS*coef;

for n=1:NCS
  vv=v(IDCASE==n,1);
  vy=v(IDCASE==n & IDDEP==1,1);
%Create gradient here
  pp=exp(vv)./sum(exp(vv),1);   %Prob for each alt
  gg=IDDEP(IDCASE==n,1)-pp;
  g(n,:)=gg'*VARS(IDCASE==n,:);
%Done with gradient
  vv=vv-repmat(vy,size(vv,1),1);
  p(n,1)=1/sum(exp(vv));
end

p=max(p,0.00000001); %As a precaution

ll=-sum(log(p),1);  %Negative since neg of ll is minimized
gr=-sum(g)';        %Transpose since fminunc needs col vector